1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2001 Daniel Hartmeier 5 * Copyright (c) 2002 - 2008 Henning Brauer 6 * Copyright (c) 2012 Gleb Smirnoff <glebius@FreeBSD.org> 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 13 * - Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * - Redistributions in binary form must reproduce the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer in the documentation and/or other materials provided 18 * with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 28 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 30 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 31 * POSSIBILITY OF SUCH DAMAGE. 32 * 33 * Effort sponsored in part by the Defense Advanced Research Projects 34 * Agency (DARPA) and Air Force Research Laboratory, Air Force 35 * Materiel Command, USAF, under agreement number F30602-01-2-0537. 36 * 37 * $OpenBSD: pf.c,v 1.634 2009/02/27 12:37:45 henning Exp $ 38 */ 39 40 #include <sys/cdefs.h> 41 __FBSDID("$FreeBSD$"); 42 43 #include "opt_bpf.h" 44 #include "opt_inet.h" 45 #include "opt_inet6.h" 46 #include "opt_pf.h" 47 #include "opt_sctp.h" 48 49 #include <sys/param.h> 50 #include <sys/bus.h> 51 #include <sys/endian.h> 52 #include <sys/gsb_crc32.h> 53 #include <sys/hash.h> 54 #include <sys/interrupt.h> 55 #include <sys/kernel.h> 56 #include <sys/kthread.h> 57 #include <sys/limits.h> 58 #include <sys/mbuf.h> 59 #include <sys/md5.h> 60 #include <sys/random.h> 61 #include <sys/refcount.h> 62 #include <sys/sdt.h> 63 #include <sys/socket.h> 64 #include <sys/sysctl.h> 65 #include <sys/taskqueue.h> 66 #include <sys/ucred.h> 67 68 #include <net/if.h> 69 #include <net/if_var.h> 70 #include <net/if_types.h> 71 #include <net/if_vlan_var.h> 72 #include <net/route.h> 73 #include <net/route/nhop.h> 74 #include <net/vnet.h> 75 76 #include <net/pfil.h> 77 #include <net/pfvar.h> 78 #include <net/if_pflog.h> 79 #include <net/if_pfsync.h> 80 81 #include <netinet/in_pcb.h> 82 #include <netinet/in_var.h> 83 #include <netinet/in_fib.h> 84 #include <netinet/ip.h> 85 #include <netinet/ip_fw.h> 86 #include <netinet/ip_icmp.h> 87 #include <netinet/icmp_var.h> 88 #include <netinet/ip_var.h> 89 #include <netinet/tcp.h> 90 #include <netinet/tcp_fsm.h> 91 #include <netinet/tcp_seq.h> 92 #include <netinet/tcp_timer.h> 93 #include <netinet/tcp_var.h> 94 #include <netinet/udp.h> 95 #include <netinet/udp_var.h> 96 97 /* dummynet */ 98 #include <netinet/ip_dummynet.h> 99 #include <netinet/ip_fw.h> 100 #include <netpfil/ipfw/dn_heap.h> 101 #include <netpfil/ipfw/ip_fw_private.h> 102 #include <netpfil/ipfw/ip_dn_private.h> 103 104 #ifdef INET6 105 #include <netinet/ip6.h> 106 #include <netinet/icmp6.h> 107 #include <netinet6/nd6.h> 108 #include <netinet6/ip6_var.h> 109 #include <netinet6/in6_pcb.h> 110 #include <netinet6/in6_fib.h> 111 #include <netinet6/scope6_var.h> 112 #endif /* INET6 */ 113 114 #if defined(SCTP) || defined(SCTP_SUPPORT) 115 #include <netinet/sctp_crc32.h> 116 #endif 117 118 #include <machine/in_cksum.h> 119 #include <security/mac/mac_framework.h> 120 121 #define DPFPRINTF(n, x) if (V_pf_status.debug >= (n)) printf x 122 123 SDT_PROVIDER_DEFINE(pf); 124 SDT_PROBE_DEFINE4(pf, ip, test, done, "int", "int", "struct pf_krule *", 125 "struct pf_kstate *"); 126 SDT_PROBE_DEFINE4(pf, ip, test6, done, "int", "int", "struct pf_krule *", 127 "struct pf_kstate *"); 128 SDT_PROBE_DEFINE5(pf, ip, state, lookup, "struct pfi_kkif *", 129 "struct pf_state_key_cmp *", "int", "struct pf_pdesc *", 130 "struct pf_kstate *"); 131 132 /* 133 * Global variables 134 */ 135 136 /* state tables */ 137 VNET_DEFINE(struct pf_altqqueue, pf_altqs[4]); 138 VNET_DEFINE(struct pf_kpalist, pf_pabuf); 139 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_active); 140 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_active); 141 VNET_DEFINE(struct pf_altqqueue *, pf_altqs_inactive); 142 VNET_DEFINE(struct pf_altqqueue *, pf_altq_ifs_inactive); 143 VNET_DEFINE(struct pf_kstatus, pf_status); 144 145 VNET_DEFINE(u_int32_t, ticket_altqs_active); 146 VNET_DEFINE(u_int32_t, ticket_altqs_inactive); 147 VNET_DEFINE(int, altqs_inactive_open); 148 VNET_DEFINE(u_int32_t, ticket_pabuf); 149 150 VNET_DEFINE(MD5_CTX, pf_tcp_secret_ctx); 151 #define V_pf_tcp_secret_ctx VNET(pf_tcp_secret_ctx) 152 VNET_DEFINE(u_char, pf_tcp_secret[16]); 153 #define V_pf_tcp_secret VNET(pf_tcp_secret) 154 VNET_DEFINE(int, pf_tcp_secret_init); 155 #define V_pf_tcp_secret_init VNET(pf_tcp_secret_init) 156 VNET_DEFINE(int, pf_tcp_iss_off); 157 #define V_pf_tcp_iss_off VNET(pf_tcp_iss_off) 158 VNET_DECLARE(int, pf_vnet_active); 159 #define V_pf_vnet_active VNET(pf_vnet_active) 160 161 VNET_DEFINE_STATIC(uint32_t, pf_purge_idx); 162 #define V_pf_purge_idx VNET(pf_purge_idx) 163 164 #ifdef PF_WANT_32_TO_64_COUNTER 165 VNET_DEFINE_STATIC(uint32_t, pf_counter_periodic_iter); 166 #define V_pf_counter_periodic_iter VNET(pf_counter_periodic_iter) 167 168 VNET_DEFINE(struct allrulelist_head, pf_allrulelist); 169 VNET_DEFINE(size_t, pf_allrulecount); 170 VNET_DEFINE(struct pf_krule *, pf_rulemarker); 171 #endif 172 173 /* 174 * Queue for pf_intr() sends. 175 */ 176 static MALLOC_DEFINE(M_PFTEMP, "pf_temp", "pf(4) temporary allocations"); 177 struct pf_send_entry { 178 STAILQ_ENTRY(pf_send_entry) pfse_next; 179 struct mbuf *pfse_m; 180 enum { 181 PFSE_IP, 182 PFSE_IP6, 183 PFSE_ICMP, 184 PFSE_ICMP6, 185 } pfse_type; 186 struct { 187 int type; 188 int code; 189 int mtu; 190 } icmpopts; 191 }; 192 193 STAILQ_HEAD(pf_send_head, pf_send_entry); 194 VNET_DEFINE_STATIC(struct pf_send_head, pf_sendqueue); 195 #define V_pf_sendqueue VNET(pf_sendqueue) 196 197 static struct mtx_padalign pf_sendqueue_mtx; 198 MTX_SYSINIT(pf_sendqueue_mtx, &pf_sendqueue_mtx, "pf send queue", MTX_DEF); 199 #define PF_SENDQ_LOCK() mtx_lock(&pf_sendqueue_mtx) 200 #define PF_SENDQ_UNLOCK() mtx_unlock(&pf_sendqueue_mtx) 201 202 /* 203 * Queue for pf_overload_task() tasks. 204 */ 205 struct pf_overload_entry { 206 SLIST_ENTRY(pf_overload_entry) next; 207 struct pf_addr addr; 208 sa_family_t af; 209 uint8_t dir; 210 struct pf_krule *rule; 211 }; 212 213 SLIST_HEAD(pf_overload_head, pf_overload_entry); 214 VNET_DEFINE_STATIC(struct pf_overload_head, pf_overloadqueue); 215 #define V_pf_overloadqueue VNET(pf_overloadqueue) 216 VNET_DEFINE_STATIC(struct task, pf_overloadtask); 217 #define V_pf_overloadtask VNET(pf_overloadtask) 218 219 static struct mtx_padalign pf_overloadqueue_mtx; 220 MTX_SYSINIT(pf_overloadqueue_mtx, &pf_overloadqueue_mtx, 221 "pf overload/flush queue", MTX_DEF); 222 #define PF_OVERLOADQ_LOCK() mtx_lock(&pf_overloadqueue_mtx) 223 #define PF_OVERLOADQ_UNLOCK() mtx_unlock(&pf_overloadqueue_mtx) 224 225 VNET_DEFINE(struct pf_krulequeue, pf_unlinked_rules); 226 struct mtx_padalign pf_unlnkdrules_mtx; 227 MTX_SYSINIT(pf_unlnkdrules_mtx, &pf_unlnkdrules_mtx, "pf unlinked rules", 228 MTX_DEF); 229 230 struct mtx_padalign pf_table_stats_lock; 231 MTX_SYSINIT(pf_table_stats_lock, &pf_table_stats_lock, "pf table stats", 232 MTX_DEF); 233 234 VNET_DEFINE_STATIC(uma_zone_t, pf_sources_z); 235 #define V_pf_sources_z VNET(pf_sources_z) 236 uma_zone_t pf_mtag_z; 237 VNET_DEFINE(uma_zone_t, pf_state_z); 238 VNET_DEFINE(uma_zone_t, pf_state_key_z); 239 240 VNET_DEFINE(uint64_t, pf_stateid[MAXCPU]); 241 #define PFID_CPUBITS 8 242 #define PFID_CPUSHIFT (sizeof(uint64_t) * NBBY - PFID_CPUBITS) 243 #define PFID_CPUMASK ((uint64_t)((1 << PFID_CPUBITS) - 1) << PFID_CPUSHIFT) 244 #define PFID_MAXID (~PFID_CPUMASK) 245 CTASSERT((1 << PFID_CPUBITS) >= MAXCPU); 246 247 static void pf_src_tree_remove_state(struct pf_kstate *); 248 static void pf_init_threshold(struct pf_threshold *, u_int32_t, 249 u_int32_t); 250 static void pf_add_threshold(struct pf_threshold *); 251 static int pf_check_threshold(struct pf_threshold *); 252 253 static void pf_change_ap(struct mbuf *, struct pf_addr *, u_int16_t *, 254 u_int16_t *, u_int16_t *, struct pf_addr *, 255 u_int16_t, u_int8_t, sa_family_t); 256 static int pf_modulate_sack(struct mbuf *, int, struct pf_pdesc *, 257 struct tcphdr *, struct pf_state_peer *); 258 static void pf_change_icmp(struct pf_addr *, u_int16_t *, 259 struct pf_addr *, struct pf_addr *, u_int16_t, 260 u_int16_t *, u_int16_t *, u_int16_t *, 261 u_int16_t *, u_int8_t, sa_family_t); 262 static void pf_send_icmp(struct mbuf *, u_int8_t, u_int8_t, 263 sa_family_t, struct pf_krule *); 264 static void pf_detach_state(struct pf_kstate *); 265 static int pf_state_key_attach(struct pf_state_key *, 266 struct pf_state_key *, struct pf_kstate *); 267 static void pf_state_key_detach(struct pf_kstate *, int); 268 static int pf_state_key_ctor(void *, int, void *, int); 269 static u_int32_t pf_tcp_iss(struct pf_pdesc *); 270 void pf_rule_to_actions(struct pf_krule *, 271 struct pf_rule_actions *); 272 static int pf_test_rule(struct pf_krule **, struct pf_kstate **, 273 int, struct pfi_kkif *, struct mbuf *, int, 274 struct pf_pdesc *, struct pf_krule **, 275 struct pf_kruleset **, struct inpcb *); 276 static int pf_create_state(struct pf_krule *, struct pf_krule *, 277 struct pf_krule *, struct pf_pdesc *, 278 struct pf_ksrc_node *, struct pf_state_key *, 279 struct pf_state_key *, struct mbuf *, int, 280 u_int16_t, u_int16_t, int *, struct pfi_kkif *, 281 struct pf_kstate **, int, u_int16_t, u_int16_t, 282 int); 283 static int pf_test_fragment(struct pf_krule **, int, 284 struct pfi_kkif *, struct mbuf *, void *, 285 struct pf_pdesc *, struct pf_krule **, 286 struct pf_kruleset **); 287 static int pf_tcp_track_full(struct pf_kstate **, 288 struct pfi_kkif *, struct mbuf *, int, 289 struct pf_pdesc *, u_short *, int *); 290 static int pf_tcp_track_sloppy(struct pf_kstate **, 291 struct pf_pdesc *, u_short *); 292 static int pf_test_state_tcp(struct pf_kstate **, int, 293 struct pfi_kkif *, struct mbuf *, int, 294 void *, struct pf_pdesc *, u_short *); 295 static int pf_test_state_udp(struct pf_kstate **, int, 296 struct pfi_kkif *, struct mbuf *, int, 297 void *, struct pf_pdesc *); 298 static int pf_test_state_icmp(struct pf_kstate **, int, 299 struct pfi_kkif *, struct mbuf *, int, 300 void *, struct pf_pdesc *, u_short *); 301 static int pf_test_state_other(struct pf_kstate **, int, 302 struct pfi_kkif *, struct mbuf *, struct pf_pdesc *); 303 static u_int16_t pf_calc_mss(struct pf_addr *, sa_family_t, 304 int, u_int16_t); 305 static int pf_check_proto_cksum(struct mbuf *, int, int, 306 u_int8_t, sa_family_t); 307 static void pf_print_state_parts(struct pf_kstate *, 308 struct pf_state_key *, struct pf_state_key *); 309 static int pf_addr_wrap_neq(struct pf_addr_wrap *, 310 struct pf_addr_wrap *); 311 static void pf_patch_8(struct mbuf *, u_int16_t *, u_int8_t *, u_int8_t, 312 bool, u_int8_t); 313 static struct pf_kstate *pf_find_state(struct pfi_kkif *, 314 struct pf_state_key_cmp *, u_int); 315 static int pf_src_connlimit(struct pf_kstate **); 316 static void pf_overload_task(void *v, int pending); 317 static int pf_insert_src_node(struct pf_ksrc_node **, 318 struct pf_krule *, struct pf_addr *, sa_family_t); 319 static u_int pf_purge_expired_states(u_int, int); 320 static void pf_purge_unlinked_rules(void); 321 static int pf_mtag_uminit(void *, int, int); 322 static void pf_mtag_free(struct m_tag *); 323 #ifdef INET 324 static void pf_route(struct mbuf **, struct pf_krule *, int, 325 struct ifnet *, struct pf_kstate *, 326 struct pf_pdesc *, struct inpcb *); 327 #endif /* INET */ 328 #ifdef INET6 329 static void pf_change_a6(struct pf_addr *, u_int16_t *, 330 struct pf_addr *, u_int8_t); 331 static void pf_route6(struct mbuf **, struct pf_krule *, int, 332 struct ifnet *, struct pf_kstate *, 333 struct pf_pdesc *, struct inpcb *); 334 #endif /* INET6 */ 335 static __inline void pf_set_protostate(struct pf_kstate *, int, u_int8_t); 336 337 int in4_cksum(struct mbuf *m, u_int8_t nxt, int off, int len); 338 339 extern int pf_end_threads; 340 extern struct proc *pf_purge_proc; 341 342 VNET_DEFINE(struct pf_limit, pf_limits[PF_LIMIT_MAX]); 343 344 #define PACKET_LOOPED(pd) ((pd)->pf_mtag && \ 345 (pd)->pf_mtag->flags & PF_PACKET_LOOPED) 346 347 #define STATE_LOOKUP(i, k, d, s, pd) \ 348 do { \ 349 (s) = pf_find_state((i), (k), (d)); \ 350 SDT_PROBE5(pf, ip, state, lookup, i, k, d, pd, (s)); \ 351 if ((s) == NULL) \ 352 return (PF_DROP); \ 353 if (PACKET_LOOPED(pd)) \ 354 return (PF_PASS); \ 355 } while (0) 356 357 #define BOUND_IFACE(r, k) \ 358 ((r)->rule_flag & PFRULE_IFBOUND) ? (k) : V_pfi_all 359 360 #define STATE_INC_COUNTERS(s) \ 361 do { \ 362 counter_u64_add(s->rule.ptr->states_cur, 1); \ 363 counter_u64_add(s->rule.ptr->states_tot, 1); \ 364 if (s->anchor.ptr != NULL) { \ 365 counter_u64_add(s->anchor.ptr->states_cur, 1); \ 366 counter_u64_add(s->anchor.ptr->states_tot, 1); \ 367 } \ 368 if (s->nat_rule.ptr != NULL) { \ 369 counter_u64_add(s->nat_rule.ptr->states_cur, 1);\ 370 counter_u64_add(s->nat_rule.ptr->states_tot, 1);\ 371 } \ 372 } while (0) 373 374 #define STATE_DEC_COUNTERS(s) \ 375 do { \ 376 if (s->nat_rule.ptr != NULL) \ 377 counter_u64_add(s->nat_rule.ptr->states_cur, -1);\ 378 if (s->anchor.ptr != NULL) \ 379 counter_u64_add(s->anchor.ptr->states_cur, -1); \ 380 counter_u64_add(s->rule.ptr->states_cur, -1); \ 381 } while (0) 382 383 MALLOC_DEFINE(M_PFHASH, "pf_hash", "pf(4) hash header structures"); 384 VNET_DEFINE(struct pf_keyhash *, pf_keyhash); 385 VNET_DEFINE(struct pf_idhash *, pf_idhash); 386 VNET_DEFINE(struct pf_srchash *, pf_srchash); 387 388 SYSCTL_NODE(_net, OID_AUTO, pf, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 389 "pf(4)"); 390 391 u_long pf_hashmask; 392 u_long pf_srchashmask; 393 static u_long pf_hashsize; 394 static u_long pf_srchashsize; 395 u_long pf_ioctl_maxcount = 65535; 396 397 SYSCTL_ULONG(_net_pf, OID_AUTO, states_hashsize, CTLFLAG_RDTUN, 398 &pf_hashsize, 0, "Size of pf(4) states hashtable"); 399 SYSCTL_ULONG(_net_pf, OID_AUTO, source_nodes_hashsize, CTLFLAG_RDTUN, 400 &pf_srchashsize, 0, "Size of pf(4) source nodes hashtable"); 401 SYSCTL_ULONG(_net_pf, OID_AUTO, request_maxcount, CTLFLAG_RWTUN, 402 &pf_ioctl_maxcount, 0, "Maximum number of tables, addresses, ... in a single ioctl() call"); 403 404 VNET_DEFINE(void *, pf_swi_cookie); 405 VNET_DEFINE(struct intr_event *, pf_swi_ie); 406 407 VNET_DEFINE(uint32_t, pf_hashseed); 408 #define V_pf_hashseed VNET(pf_hashseed) 409 410 int 411 pf_addr_cmp(struct pf_addr *a, struct pf_addr *b, sa_family_t af) 412 { 413 414 switch (af) { 415 #ifdef INET 416 case AF_INET: 417 if (a->addr32[0] > b->addr32[0]) 418 return (1); 419 if (a->addr32[0] < b->addr32[0]) 420 return (-1); 421 break; 422 #endif /* INET */ 423 #ifdef INET6 424 case AF_INET6: 425 if (a->addr32[3] > b->addr32[3]) 426 return (1); 427 if (a->addr32[3] < b->addr32[3]) 428 return (-1); 429 if (a->addr32[2] > b->addr32[2]) 430 return (1); 431 if (a->addr32[2] < b->addr32[2]) 432 return (-1); 433 if (a->addr32[1] > b->addr32[1]) 434 return (1); 435 if (a->addr32[1] < b->addr32[1]) 436 return (-1); 437 if (a->addr32[0] > b->addr32[0]) 438 return (1); 439 if (a->addr32[0] < b->addr32[0]) 440 return (-1); 441 break; 442 #endif /* INET6 */ 443 default: 444 panic("%s: unknown address family %u", __func__, af); 445 } 446 return (0); 447 } 448 449 static __inline uint32_t 450 pf_hashkey(struct pf_state_key *sk) 451 { 452 uint32_t h; 453 454 h = murmur3_32_hash32((uint32_t *)sk, 455 sizeof(struct pf_state_key_cmp)/sizeof(uint32_t), 456 V_pf_hashseed); 457 458 return (h & pf_hashmask); 459 } 460 461 static __inline uint32_t 462 pf_hashsrc(struct pf_addr *addr, sa_family_t af) 463 { 464 uint32_t h; 465 466 switch (af) { 467 case AF_INET: 468 h = murmur3_32_hash32((uint32_t *)&addr->v4, 469 sizeof(addr->v4)/sizeof(uint32_t), V_pf_hashseed); 470 break; 471 case AF_INET6: 472 h = murmur3_32_hash32((uint32_t *)&addr->v6, 473 sizeof(addr->v6)/sizeof(uint32_t), V_pf_hashseed); 474 break; 475 default: 476 panic("%s: unknown address family %u", __func__, af); 477 } 478 479 return (h & pf_srchashmask); 480 } 481 482 #ifdef ALTQ 483 static int 484 pf_state_hash(struct pf_kstate *s) 485 { 486 u_int32_t hv = (intptr_t)s / sizeof(*s); 487 488 hv ^= crc32(&s->src, sizeof(s->src)); 489 hv ^= crc32(&s->dst, sizeof(s->dst)); 490 if (hv == 0) 491 hv = 1; 492 return (hv); 493 } 494 #endif 495 496 static __inline void 497 pf_set_protostate(struct pf_kstate *s, int which, u_int8_t newstate) 498 { 499 if (which == PF_PEER_DST || which == PF_PEER_BOTH) 500 s->dst.state = newstate; 501 if (which == PF_PEER_DST) 502 return; 503 504 s->src.state = newstate; 505 } 506 507 #ifdef INET6 508 void 509 pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) 510 { 511 switch (af) { 512 #ifdef INET 513 case AF_INET: 514 dst->addr32[0] = src->addr32[0]; 515 break; 516 #endif /* INET */ 517 case AF_INET6: 518 dst->addr32[0] = src->addr32[0]; 519 dst->addr32[1] = src->addr32[1]; 520 dst->addr32[2] = src->addr32[2]; 521 dst->addr32[3] = src->addr32[3]; 522 break; 523 } 524 } 525 #endif /* INET6 */ 526 527 static void 528 pf_init_threshold(struct pf_threshold *threshold, 529 u_int32_t limit, u_int32_t seconds) 530 { 531 threshold->limit = limit * PF_THRESHOLD_MULT; 532 threshold->seconds = seconds; 533 threshold->count = 0; 534 threshold->last = time_uptime; 535 } 536 537 static void 538 pf_add_threshold(struct pf_threshold *threshold) 539 { 540 u_int32_t t = time_uptime, diff = t - threshold->last; 541 542 if (diff >= threshold->seconds) 543 threshold->count = 0; 544 else 545 threshold->count -= threshold->count * diff / 546 threshold->seconds; 547 threshold->count += PF_THRESHOLD_MULT; 548 threshold->last = t; 549 } 550 551 static int 552 pf_check_threshold(struct pf_threshold *threshold) 553 { 554 return (threshold->count > threshold->limit); 555 } 556 557 static int 558 pf_src_connlimit(struct pf_kstate **state) 559 { 560 struct pf_overload_entry *pfoe; 561 int bad = 0; 562 563 PF_STATE_LOCK_ASSERT(*state); 564 565 (*state)->src_node->conn++; 566 (*state)->src.tcp_est = 1; 567 pf_add_threshold(&(*state)->src_node->conn_rate); 568 569 if ((*state)->rule.ptr->max_src_conn && 570 (*state)->rule.ptr->max_src_conn < 571 (*state)->src_node->conn) { 572 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONN], 1); 573 bad++; 574 } 575 576 if ((*state)->rule.ptr->max_src_conn_rate.limit && 577 pf_check_threshold(&(*state)->src_node->conn_rate)) { 578 counter_u64_add(V_pf_status.lcounters[LCNT_SRCCONNRATE], 1); 579 bad++; 580 } 581 582 if (!bad) 583 return (0); 584 585 /* Kill this state. */ 586 (*state)->timeout = PFTM_PURGE; 587 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 588 589 if ((*state)->rule.ptr->overload_tbl == NULL) 590 return (1); 591 592 /* Schedule overloading and flushing task. */ 593 pfoe = malloc(sizeof(*pfoe), M_PFTEMP, M_NOWAIT); 594 if (pfoe == NULL) 595 return (1); /* too bad :( */ 596 597 bcopy(&(*state)->src_node->addr, &pfoe->addr, sizeof(pfoe->addr)); 598 pfoe->af = (*state)->key[PF_SK_WIRE]->af; 599 pfoe->rule = (*state)->rule.ptr; 600 pfoe->dir = (*state)->direction; 601 PF_OVERLOADQ_LOCK(); 602 SLIST_INSERT_HEAD(&V_pf_overloadqueue, pfoe, next); 603 PF_OVERLOADQ_UNLOCK(); 604 taskqueue_enqueue(taskqueue_swi, &V_pf_overloadtask); 605 606 return (1); 607 } 608 609 static void 610 pf_overload_task(void *v, int pending) 611 { 612 struct pf_overload_head queue; 613 struct pfr_addr p; 614 struct pf_overload_entry *pfoe, *pfoe1; 615 uint32_t killed = 0; 616 617 CURVNET_SET((struct vnet *)v); 618 619 PF_OVERLOADQ_LOCK(); 620 queue = V_pf_overloadqueue; 621 SLIST_INIT(&V_pf_overloadqueue); 622 PF_OVERLOADQ_UNLOCK(); 623 624 bzero(&p, sizeof(p)); 625 SLIST_FOREACH(pfoe, &queue, next) { 626 counter_u64_add(V_pf_status.lcounters[LCNT_OVERLOAD_TABLE], 1); 627 if (V_pf_status.debug >= PF_DEBUG_MISC) { 628 printf("%s: blocking address ", __func__); 629 pf_print_host(&pfoe->addr, 0, pfoe->af); 630 printf("\n"); 631 } 632 633 p.pfra_af = pfoe->af; 634 switch (pfoe->af) { 635 #ifdef INET 636 case AF_INET: 637 p.pfra_net = 32; 638 p.pfra_ip4addr = pfoe->addr.v4; 639 break; 640 #endif 641 #ifdef INET6 642 case AF_INET6: 643 p.pfra_net = 128; 644 p.pfra_ip6addr = pfoe->addr.v6; 645 break; 646 #endif 647 } 648 649 PF_RULES_WLOCK(); 650 pfr_insert_kentry(pfoe->rule->overload_tbl, &p, time_second); 651 PF_RULES_WUNLOCK(); 652 } 653 654 /* 655 * Remove those entries, that don't need flushing. 656 */ 657 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 658 if (pfoe->rule->flush == 0) { 659 SLIST_REMOVE(&queue, pfoe, pf_overload_entry, next); 660 free(pfoe, M_PFTEMP); 661 } else 662 counter_u64_add( 663 V_pf_status.lcounters[LCNT_OVERLOAD_FLUSH], 1); 664 665 /* If nothing to flush, return. */ 666 if (SLIST_EMPTY(&queue)) { 667 CURVNET_RESTORE(); 668 return; 669 } 670 671 for (int i = 0; i <= pf_hashmask; i++) { 672 struct pf_idhash *ih = &V_pf_idhash[i]; 673 struct pf_state_key *sk; 674 struct pf_kstate *s; 675 676 PF_HASHROW_LOCK(ih); 677 LIST_FOREACH(s, &ih->states, entry) { 678 sk = s->key[PF_SK_WIRE]; 679 SLIST_FOREACH(pfoe, &queue, next) 680 if (sk->af == pfoe->af && 681 ((pfoe->rule->flush & PF_FLUSH_GLOBAL) || 682 pfoe->rule == s->rule.ptr) && 683 ((pfoe->dir == PF_OUT && 684 PF_AEQ(&pfoe->addr, &sk->addr[1], sk->af)) || 685 (pfoe->dir == PF_IN && 686 PF_AEQ(&pfoe->addr, &sk->addr[0], sk->af)))) { 687 s->timeout = PFTM_PURGE; 688 pf_set_protostate(s, PF_PEER_BOTH, TCPS_CLOSED); 689 killed++; 690 } 691 } 692 PF_HASHROW_UNLOCK(ih); 693 } 694 SLIST_FOREACH_SAFE(pfoe, &queue, next, pfoe1) 695 free(pfoe, M_PFTEMP); 696 if (V_pf_status.debug >= PF_DEBUG_MISC) 697 printf("%s: %u states killed", __func__, killed); 698 699 CURVNET_RESTORE(); 700 } 701 702 /* 703 * Can return locked on failure, so that we can consistently 704 * allocate and insert a new one. 705 */ 706 struct pf_ksrc_node * 707 pf_find_src_node(struct pf_addr *src, struct pf_krule *rule, sa_family_t af, 708 int returnlocked) 709 { 710 struct pf_srchash *sh; 711 struct pf_ksrc_node *n; 712 713 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_SEARCH], 1); 714 715 sh = &V_pf_srchash[pf_hashsrc(src, af)]; 716 PF_HASHROW_LOCK(sh); 717 LIST_FOREACH(n, &sh->nodes, entry) 718 if (n->rule.ptr == rule && n->af == af && 719 ((af == AF_INET && n->addr.v4.s_addr == src->v4.s_addr) || 720 (af == AF_INET6 && bcmp(&n->addr, src, sizeof(*src)) == 0))) 721 break; 722 if (n != NULL) { 723 n->states++; 724 PF_HASHROW_UNLOCK(sh); 725 } else if (returnlocked == 0) 726 PF_HASHROW_UNLOCK(sh); 727 728 return (n); 729 } 730 731 static void 732 pf_free_src_node(struct pf_ksrc_node *sn) 733 { 734 735 for (int i = 0; i < 2; i++) { 736 counter_u64_free(sn->bytes[i]); 737 counter_u64_free(sn->packets[i]); 738 } 739 uma_zfree(V_pf_sources_z, sn); 740 } 741 742 static int 743 pf_insert_src_node(struct pf_ksrc_node **sn, struct pf_krule *rule, 744 struct pf_addr *src, sa_family_t af) 745 { 746 747 KASSERT((rule->rule_flag & PFRULE_SRCTRACK || 748 rule->rpool.opts & PF_POOL_STICKYADDR), 749 ("%s for non-tracking rule %p", __func__, rule)); 750 751 if (*sn == NULL) 752 *sn = pf_find_src_node(src, rule, af, 1); 753 754 if (*sn == NULL) { 755 struct pf_srchash *sh = &V_pf_srchash[pf_hashsrc(src, af)]; 756 757 PF_HASHROW_ASSERT(sh); 758 759 if (!rule->max_src_nodes || 760 counter_u64_fetch(rule->src_nodes) < rule->max_src_nodes) 761 (*sn) = uma_zalloc(V_pf_sources_z, M_NOWAIT | M_ZERO); 762 else 763 counter_u64_add(V_pf_status.lcounters[LCNT_SRCNODES], 764 1); 765 if ((*sn) == NULL) { 766 PF_HASHROW_UNLOCK(sh); 767 return (-1); 768 } 769 770 for (int i = 0; i < 2; i++) { 771 (*sn)->bytes[i] = counter_u64_alloc(M_NOWAIT); 772 (*sn)->packets[i] = counter_u64_alloc(M_NOWAIT); 773 774 if ((*sn)->bytes[i] == NULL || (*sn)->packets[i] == NULL) { 775 pf_free_src_node(*sn); 776 PF_HASHROW_UNLOCK(sh); 777 return (-1); 778 } 779 } 780 781 pf_init_threshold(&(*sn)->conn_rate, 782 rule->max_src_conn_rate.limit, 783 rule->max_src_conn_rate.seconds); 784 785 (*sn)->af = af; 786 (*sn)->rule.ptr = rule; 787 PF_ACPY(&(*sn)->addr, src, af); 788 LIST_INSERT_HEAD(&sh->nodes, *sn, entry); 789 (*sn)->creation = time_uptime; 790 (*sn)->ruletype = rule->action; 791 (*sn)->states = 1; 792 if ((*sn)->rule.ptr != NULL) 793 counter_u64_add((*sn)->rule.ptr->src_nodes, 1); 794 PF_HASHROW_UNLOCK(sh); 795 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_INSERT], 1); 796 } else { 797 if (rule->max_src_states && 798 (*sn)->states >= rule->max_src_states) { 799 counter_u64_add(V_pf_status.lcounters[LCNT_SRCSTATES], 800 1); 801 return (-1); 802 } 803 } 804 return (0); 805 } 806 807 void 808 pf_unlink_src_node(struct pf_ksrc_node *src) 809 { 810 811 PF_HASHROW_ASSERT(&V_pf_srchash[pf_hashsrc(&src->addr, src->af)]); 812 LIST_REMOVE(src, entry); 813 if (src->rule.ptr) 814 counter_u64_add(src->rule.ptr->src_nodes, -1); 815 } 816 817 u_int 818 pf_free_src_nodes(struct pf_ksrc_node_list *head) 819 { 820 struct pf_ksrc_node *sn, *tmp; 821 u_int count = 0; 822 823 LIST_FOREACH_SAFE(sn, head, entry, tmp) { 824 pf_free_src_node(sn); 825 count++; 826 } 827 828 counter_u64_add(V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], count); 829 830 return (count); 831 } 832 833 void 834 pf_mtag_initialize() 835 { 836 837 pf_mtag_z = uma_zcreate("pf mtags", sizeof(struct m_tag) + 838 sizeof(struct pf_mtag), NULL, NULL, pf_mtag_uminit, NULL, 839 UMA_ALIGN_PTR, 0); 840 } 841 842 /* Per-vnet data storage structures initialization. */ 843 void 844 pf_initialize() 845 { 846 struct pf_keyhash *kh; 847 struct pf_idhash *ih; 848 struct pf_srchash *sh; 849 u_int i; 850 851 if (pf_hashsize == 0 || !powerof2(pf_hashsize)) 852 pf_hashsize = PF_HASHSIZ; 853 if (pf_srchashsize == 0 || !powerof2(pf_srchashsize)) 854 pf_srchashsize = PF_SRCHASHSIZ; 855 856 V_pf_hashseed = arc4random(); 857 858 /* States and state keys storage. */ 859 V_pf_state_z = uma_zcreate("pf states", sizeof(struct pf_kstate), 860 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 861 V_pf_limits[PF_LIMIT_STATES].zone = V_pf_state_z; 862 uma_zone_set_max(V_pf_state_z, PFSTATE_HIWAT); 863 uma_zone_set_warning(V_pf_state_z, "PF states limit reached"); 864 865 V_pf_state_key_z = uma_zcreate("pf state keys", 866 sizeof(struct pf_state_key), pf_state_key_ctor, NULL, NULL, NULL, 867 UMA_ALIGN_PTR, 0); 868 869 V_pf_keyhash = mallocarray(pf_hashsize, sizeof(struct pf_keyhash), 870 M_PFHASH, M_NOWAIT | M_ZERO); 871 V_pf_idhash = mallocarray(pf_hashsize, sizeof(struct pf_idhash), 872 M_PFHASH, M_NOWAIT | M_ZERO); 873 if (V_pf_keyhash == NULL || V_pf_idhash == NULL) { 874 printf("pf: Unable to allocate memory for " 875 "state_hashsize %lu.\n", pf_hashsize); 876 877 free(V_pf_keyhash, M_PFHASH); 878 free(V_pf_idhash, M_PFHASH); 879 880 pf_hashsize = PF_HASHSIZ; 881 V_pf_keyhash = mallocarray(pf_hashsize, 882 sizeof(struct pf_keyhash), M_PFHASH, M_WAITOK | M_ZERO); 883 V_pf_idhash = mallocarray(pf_hashsize, 884 sizeof(struct pf_idhash), M_PFHASH, M_WAITOK | M_ZERO); 885 } 886 887 pf_hashmask = pf_hashsize - 1; 888 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 889 i++, kh++, ih++) { 890 mtx_init(&kh->lock, "pf_keyhash", NULL, MTX_DEF | MTX_DUPOK); 891 mtx_init(&ih->lock, "pf_idhash", NULL, MTX_DEF); 892 } 893 894 /* Source nodes. */ 895 V_pf_sources_z = uma_zcreate("pf source nodes", 896 sizeof(struct pf_ksrc_node), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 897 0); 898 V_pf_limits[PF_LIMIT_SRC_NODES].zone = V_pf_sources_z; 899 uma_zone_set_max(V_pf_sources_z, PFSNODE_HIWAT); 900 uma_zone_set_warning(V_pf_sources_z, "PF source nodes limit reached"); 901 902 V_pf_srchash = mallocarray(pf_srchashsize, 903 sizeof(struct pf_srchash), M_PFHASH, M_NOWAIT | M_ZERO); 904 if (V_pf_srchash == NULL) { 905 printf("pf: Unable to allocate memory for " 906 "source_hashsize %lu.\n", pf_srchashsize); 907 908 pf_srchashsize = PF_SRCHASHSIZ; 909 V_pf_srchash = mallocarray(pf_srchashsize, 910 sizeof(struct pf_srchash), M_PFHASH, M_WAITOK | M_ZERO); 911 } 912 913 pf_srchashmask = pf_srchashsize - 1; 914 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) 915 mtx_init(&sh->lock, "pf_srchash", NULL, MTX_DEF); 916 917 /* ALTQ */ 918 TAILQ_INIT(&V_pf_altqs[0]); 919 TAILQ_INIT(&V_pf_altqs[1]); 920 TAILQ_INIT(&V_pf_altqs[2]); 921 TAILQ_INIT(&V_pf_altqs[3]); 922 TAILQ_INIT(&V_pf_pabuf); 923 V_pf_altqs_active = &V_pf_altqs[0]; 924 V_pf_altq_ifs_active = &V_pf_altqs[1]; 925 V_pf_altqs_inactive = &V_pf_altqs[2]; 926 V_pf_altq_ifs_inactive = &V_pf_altqs[3]; 927 928 /* Send & overload+flush queues. */ 929 STAILQ_INIT(&V_pf_sendqueue); 930 SLIST_INIT(&V_pf_overloadqueue); 931 TASK_INIT(&V_pf_overloadtask, 0, pf_overload_task, curvnet); 932 933 /* Unlinked, but may be referenced rules. */ 934 TAILQ_INIT(&V_pf_unlinked_rules); 935 } 936 937 void 938 pf_mtag_cleanup() 939 { 940 941 uma_zdestroy(pf_mtag_z); 942 } 943 944 void 945 pf_cleanup() 946 { 947 struct pf_keyhash *kh; 948 struct pf_idhash *ih; 949 struct pf_srchash *sh; 950 struct pf_send_entry *pfse, *next; 951 u_int i; 952 953 for (i = 0, kh = V_pf_keyhash, ih = V_pf_idhash; i <= pf_hashmask; 954 i++, kh++, ih++) { 955 KASSERT(LIST_EMPTY(&kh->keys), ("%s: key hash not empty", 956 __func__)); 957 KASSERT(LIST_EMPTY(&ih->states), ("%s: id hash not empty", 958 __func__)); 959 mtx_destroy(&kh->lock); 960 mtx_destroy(&ih->lock); 961 } 962 free(V_pf_keyhash, M_PFHASH); 963 free(V_pf_idhash, M_PFHASH); 964 965 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 966 KASSERT(LIST_EMPTY(&sh->nodes), 967 ("%s: source node hash not empty", __func__)); 968 mtx_destroy(&sh->lock); 969 } 970 free(V_pf_srchash, M_PFHASH); 971 972 STAILQ_FOREACH_SAFE(pfse, &V_pf_sendqueue, pfse_next, next) { 973 m_freem(pfse->pfse_m); 974 free(pfse, M_PFTEMP); 975 } 976 977 uma_zdestroy(V_pf_sources_z); 978 uma_zdestroy(V_pf_state_z); 979 uma_zdestroy(V_pf_state_key_z); 980 } 981 982 static int 983 pf_mtag_uminit(void *mem, int size, int how) 984 { 985 struct m_tag *t; 986 987 t = (struct m_tag *)mem; 988 t->m_tag_cookie = MTAG_ABI_COMPAT; 989 t->m_tag_id = PACKET_TAG_PF; 990 t->m_tag_len = sizeof(struct pf_mtag); 991 t->m_tag_free = pf_mtag_free; 992 993 return (0); 994 } 995 996 static void 997 pf_mtag_free(struct m_tag *t) 998 { 999 1000 uma_zfree(pf_mtag_z, t); 1001 } 1002 1003 struct pf_mtag * 1004 pf_get_mtag(struct mbuf *m) 1005 { 1006 struct m_tag *mtag; 1007 1008 if ((mtag = m_tag_find(m, PACKET_TAG_PF, NULL)) != NULL) 1009 return ((struct pf_mtag *)(mtag + 1)); 1010 1011 mtag = uma_zalloc(pf_mtag_z, M_NOWAIT); 1012 if (mtag == NULL) 1013 return (NULL); 1014 bzero(mtag + 1, sizeof(struct pf_mtag)); 1015 m_tag_prepend(m, mtag); 1016 1017 return ((struct pf_mtag *)(mtag + 1)); 1018 } 1019 1020 static int 1021 pf_state_key_attach(struct pf_state_key *skw, struct pf_state_key *sks, 1022 struct pf_kstate *s) 1023 { 1024 struct pf_keyhash *khs, *khw, *kh; 1025 struct pf_state_key *sk, *cur; 1026 struct pf_kstate *si, *olds = NULL; 1027 int idx; 1028 1029 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1030 KASSERT(s->key[PF_SK_WIRE] == NULL, ("%s: state has key", __func__)); 1031 KASSERT(s->key[PF_SK_STACK] == NULL, ("%s: state has key", __func__)); 1032 1033 /* 1034 * We need to lock hash slots of both keys. To avoid deadlock 1035 * we always lock the slot with lower address first. Unlock order 1036 * isn't important. 1037 * 1038 * We also need to lock ID hash slot before dropping key 1039 * locks. On success we return with ID hash slot locked. 1040 */ 1041 1042 if (skw == sks) { 1043 khs = khw = &V_pf_keyhash[pf_hashkey(skw)]; 1044 PF_HASHROW_LOCK(khs); 1045 } else { 1046 khs = &V_pf_keyhash[pf_hashkey(sks)]; 1047 khw = &V_pf_keyhash[pf_hashkey(skw)]; 1048 if (khs == khw) { 1049 PF_HASHROW_LOCK(khs); 1050 } else if (khs < khw) { 1051 PF_HASHROW_LOCK(khs); 1052 PF_HASHROW_LOCK(khw); 1053 } else { 1054 PF_HASHROW_LOCK(khw); 1055 PF_HASHROW_LOCK(khs); 1056 } 1057 } 1058 1059 #define KEYS_UNLOCK() do { \ 1060 if (khs != khw) { \ 1061 PF_HASHROW_UNLOCK(khs); \ 1062 PF_HASHROW_UNLOCK(khw); \ 1063 } else \ 1064 PF_HASHROW_UNLOCK(khs); \ 1065 } while (0) 1066 1067 /* 1068 * First run: start with wire key. 1069 */ 1070 sk = skw; 1071 kh = khw; 1072 idx = PF_SK_WIRE; 1073 1074 MPASS(s->lock == NULL); 1075 s->lock = &V_pf_idhash[PF_IDHASH(s)].lock; 1076 1077 keyattach: 1078 LIST_FOREACH(cur, &kh->keys, entry) 1079 if (bcmp(cur, sk, sizeof(struct pf_state_key_cmp)) == 0) 1080 break; 1081 1082 if (cur != NULL) { 1083 /* Key exists. Check for same kif, if none, add to key. */ 1084 TAILQ_FOREACH(si, &cur->states[idx], key_list[idx]) { 1085 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(si)]; 1086 1087 PF_HASHROW_LOCK(ih); 1088 if (si->kif == s->kif && 1089 si->direction == s->direction) { 1090 if (sk->proto == IPPROTO_TCP && 1091 si->src.state >= TCPS_FIN_WAIT_2 && 1092 si->dst.state >= TCPS_FIN_WAIT_2) { 1093 /* 1094 * New state matches an old >FIN_WAIT_2 1095 * state. We can't drop key hash locks, 1096 * thus we can't unlink it properly. 1097 * 1098 * As a workaround we drop it into 1099 * TCPS_CLOSED state, schedule purge 1100 * ASAP and push it into the very end 1101 * of the slot TAILQ, so that it won't 1102 * conflict with our new state. 1103 */ 1104 pf_set_protostate(si, PF_PEER_BOTH, 1105 TCPS_CLOSED); 1106 si->timeout = PFTM_PURGE; 1107 olds = si; 1108 } else { 1109 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1110 printf("pf: %s key attach " 1111 "failed on %s: ", 1112 (idx == PF_SK_WIRE) ? 1113 "wire" : "stack", 1114 s->kif->pfik_name); 1115 pf_print_state_parts(s, 1116 (idx == PF_SK_WIRE) ? 1117 sk : NULL, 1118 (idx == PF_SK_STACK) ? 1119 sk : NULL); 1120 printf(", existing: "); 1121 pf_print_state_parts(si, 1122 (idx == PF_SK_WIRE) ? 1123 sk : NULL, 1124 (idx == PF_SK_STACK) ? 1125 sk : NULL); 1126 printf("\n"); 1127 } 1128 PF_HASHROW_UNLOCK(ih); 1129 KEYS_UNLOCK(); 1130 uma_zfree(V_pf_state_key_z, sk); 1131 if (idx == PF_SK_STACK) 1132 pf_detach_state(s); 1133 return (EEXIST); /* collision! */ 1134 } 1135 } 1136 PF_HASHROW_UNLOCK(ih); 1137 } 1138 uma_zfree(V_pf_state_key_z, sk); 1139 s->key[idx] = cur; 1140 } else { 1141 LIST_INSERT_HEAD(&kh->keys, sk, entry); 1142 s->key[idx] = sk; 1143 } 1144 1145 stateattach: 1146 /* List is sorted, if-bound states before floating. */ 1147 if (s->kif == V_pfi_all) 1148 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], s, key_list[idx]); 1149 else 1150 TAILQ_INSERT_HEAD(&s->key[idx]->states[idx], s, key_list[idx]); 1151 1152 if (olds) { 1153 TAILQ_REMOVE(&s->key[idx]->states[idx], olds, key_list[idx]); 1154 TAILQ_INSERT_TAIL(&s->key[idx]->states[idx], olds, 1155 key_list[idx]); 1156 olds = NULL; 1157 } 1158 1159 /* 1160 * Attach done. See how should we (or should not?) 1161 * attach a second key. 1162 */ 1163 if (sks == skw) { 1164 s->key[PF_SK_STACK] = s->key[PF_SK_WIRE]; 1165 idx = PF_SK_STACK; 1166 sks = NULL; 1167 goto stateattach; 1168 } else if (sks != NULL) { 1169 /* 1170 * Continue attaching with stack key. 1171 */ 1172 sk = sks; 1173 kh = khs; 1174 idx = PF_SK_STACK; 1175 sks = NULL; 1176 goto keyattach; 1177 } 1178 1179 PF_STATE_LOCK(s); 1180 KEYS_UNLOCK(); 1181 1182 KASSERT(s->key[PF_SK_WIRE] != NULL && s->key[PF_SK_STACK] != NULL, 1183 ("%s failure", __func__)); 1184 1185 return (0); 1186 #undef KEYS_UNLOCK 1187 } 1188 1189 static void 1190 pf_detach_state(struct pf_kstate *s) 1191 { 1192 struct pf_state_key *sks = s->key[PF_SK_STACK]; 1193 struct pf_keyhash *kh; 1194 1195 if (sks != NULL) { 1196 kh = &V_pf_keyhash[pf_hashkey(sks)]; 1197 PF_HASHROW_LOCK(kh); 1198 if (s->key[PF_SK_STACK] != NULL) 1199 pf_state_key_detach(s, PF_SK_STACK); 1200 /* 1201 * If both point to same key, then we are done. 1202 */ 1203 if (sks == s->key[PF_SK_WIRE]) { 1204 pf_state_key_detach(s, PF_SK_WIRE); 1205 PF_HASHROW_UNLOCK(kh); 1206 return; 1207 } 1208 PF_HASHROW_UNLOCK(kh); 1209 } 1210 1211 if (s->key[PF_SK_WIRE] != NULL) { 1212 kh = &V_pf_keyhash[pf_hashkey(s->key[PF_SK_WIRE])]; 1213 PF_HASHROW_LOCK(kh); 1214 if (s->key[PF_SK_WIRE] != NULL) 1215 pf_state_key_detach(s, PF_SK_WIRE); 1216 PF_HASHROW_UNLOCK(kh); 1217 } 1218 } 1219 1220 static void 1221 pf_state_key_detach(struct pf_kstate *s, int idx) 1222 { 1223 struct pf_state_key *sk = s->key[idx]; 1224 #ifdef INVARIANTS 1225 struct pf_keyhash *kh = &V_pf_keyhash[pf_hashkey(sk)]; 1226 1227 PF_HASHROW_ASSERT(kh); 1228 #endif 1229 TAILQ_REMOVE(&sk->states[idx], s, key_list[idx]); 1230 s->key[idx] = NULL; 1231 1232 if (TAILQ_EMPTY(&sk->states[0]) && TAILQ_EMPTY(&sk->states[1])) { 1233 LIST_REMOVE(sk, entry); 1234 uma_zfree(V_pf_state_key_z, sk); 1235 } 1236 } 1237 1238 static int 1239 pf_state_key_ctor(void *mem, int size, void *arg, int flags) 1240 { 1241 struct pf_state_key *sk = mem; 1242 1243 bzero(sk, sizeof(struct pf_state_key_cmp)); 1244 TAILQ_INIT(&sk->states[PF_SK_WIRE]); 1245 TAILQ_INIT(&sk->states[PF_SK_STACK]); 1246 1247 return (0); 1248 } 1249 1250 struct pf_state_key * 1251 pf_state_key_setup(struct pf_pdesc *pd, struct pf_addr *saddr, 1252 struct pf_addr *daddr, u_int16_t sport, u_int16_t dport) 1253 { 1254 struct pf_state_key *sk; 1255 1256 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1257 if (sk == NULL) 1258 return (NULL); 1259 1260 PF_ACPY(&sk->addr[pd->sidx], saddr, pd->af); 1261 PF_ACPY(&sk->addr[pd->didx], daddr, pd->af); 1262 sk->port[pd->sidx] = sport; 1263 sk->port[pd->didx] = dport; 1264 sk->proto = pd->proto; 1265 sk->af = pd->af; 1266 1267 return (sk); 1268 } 1269 1270 struct pf_state_key * 1271 pf_state_key_clone(struct pf_state_key *orig) 1272 { 1273 struct pf_state_key *sk; 1274 1275 sk = uma_zalloc(V_pf_state_key_z, M_NOWAIT); 1276 if (sk == NULL) 1277 return (NULL); 1278 1279 bcopy(orig, sk, sizeof(struct pf_state_key_cmp)); 1280 1281 return (sk); 1282 } 1283 1284 int 1285 pf_state_insert(struct pfi_kkif *kif, struct pfi_kkif *orig_kif, 1286 struct pf_state_key *skw, struct pf_state_key *sks, struct pf_kstate *s) 1287 { 1288 struct pf_idhash *ih; 1289 struct pf_kstate *cur; 1290 int error; 1291 1292 KASSERT(TAILQ_EMPTY(&sks->states[0]) && TAILQ_EMPTY(&sks->states[1]), 1293 ("%s: sks not pristine", __func__)); 1294 KASSERT(TAILQ_EMPTY(&skw->states[0]) && TAILQ_EMPTY(&skw->states[1]), 1295 ("%s: skw not pristine", __func__)); 1296 KASSERT(s->refs == 0, ("%s: state not pristine", __func__)); 1297 1298 s->kif = kif; 1299 s->orig_kif = orig_kif; 1300 1301 if (s->id == 0 && s->creatorid == 0) { 1302 /* XXX: should be atomic, but probability of collision low */ 1303 if ((s->id = V_pf_stateid[curcpu]++) == PFID_MAXID) 1304 V_pf_stateid[curcpu] = 1; 1305 s->id |= (uint64_t )curcpu << PFID_CPUSHIFT; 1306 s->id = htobe64(s->id); 1307 s->creatorid = V_pf_status.hostid; 1308 } 1309 1310 /* Returns with ID locked on success. */ 1311 if ((error = pf_state_key_attach(skw, sks, s)) != 0) 1312 return (error); 1313 1314 ih = &V_pf_idhash[PF_IDHASH(s)]; 1315 PF_HASHROW_ASSERT(ih); 1316 LIST_FOREACH(cur, &ih->states, entry) 1317 if (cur->id == s->id && cur->creatorid == s->creatorid) 1318 break; 1319 1320 if (cur != NULL) { 1321 PF_HASHROW_UNLOCK(ih); 1322 if (V_pf_status.debug >= PF_DEBUG_MISC) { 1323 printf("pf: state ID collision: " 1324 "id: %016llx creatorid: %08x\n", 1325 (unsigned long long)be64toh(s->id), 1326 ntohl(s->creatorid)); 1327 } 1328 pf_detach_state(s); 1329 return (EEXIST); 1330 } 1331 LIST_INSERT_HEAD(&ih->states, s, entry); 1332 /* One for keys, one for ID hash. */ 1333 refcount_init(&s->refs, 2); 1334 1335 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_INSERT], 1); 1336 if (V_pfsync_insert_state_ptr != NULL) 1337 V_pfsync_insert_state_ptr(s); 1338 1339 /* Returns locked. */ 1340 return (0); 1341 } 1342 1343 /* 1344 * Find state by ID: returns with locked row on success. 1345 */ 1346 struct pf_kstate * 1347 pf_find_state_byid(uint64_t id, uint32_t creatorid) 1348 { 1349 struct pf_idhash *ih; 1350 struct pf_kstate *s; 1351 1352 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1353 1354 ih = &V_pf_idhash[(be64toh(id) % (pf_hashmask + 1))]; 1355 1356 PF_HASHROW_LOCK(ih); 1357 LIST_FOREACH(s, &ih->states, entry) 1358 if (s->id == id && s->creatorid == creatorid) 1359 break; 1360 1361 if (s == NULL) 1362 PF_HASHROW_UNLOCK(ih); 1363 1364 return (s); 1365 } 1366 1367 /* 1368 * Find state by key. 1369 * Returns with ID hash slot locked on success. 1370 */ 1371 static struct pf_kstate * 1372 pf_find_state(struct pfi_kkif *kif, struct pf_state_key_cmp *key, u_int dir) 1373 { 1374 struct pf_keyhash *kh; 1375 struct pf_state_key *sk; 1376 struct pf_kstate *s; 1377 int idx; 1378 1379 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1380 1381 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1382 1383 PF_HASHROW_LOCK(kh); 1384 LIST_FOREACH(sk, &kh->keys, entry) 1385 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1386 break; 1387 if (sk == NULL) { 1388 PF_HASHROW_UNLOCK(kh); 1389 return (NULL); 1390 } 1391 1392 idx = (dir == PF_IN ? PF_SK_WIRE : PF_SK_STACK); 1393 1394 /* List is sorted, if-bound states before floating ones. */ 1395 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) 1396 if (s->kif == V_pfi_all || s->kif == kif) { 1397 PF_STATE_LOCK(s); 1398 PF_HASHROW_UNLOCK(kh); 1399 if (__predict_false(s->timeout >= PFTM_MAX)) { 1400 /* 1401 * State is either being processed by 1402 * pf_unlink_state() in an other thread, or 1403 * is scheduled for immediate expiry. 1404 */ 1405 PF_STATE_UNLOCK(s); 1406 return (NULL); 1407 } 1408 return (s); 1409 } 1410 PF_HASHROW_UNLOCK(kh); 1411 1412 return (NULL); 1413 } 1414 1415 struct pf_kstate * 1416 pf_find_state_all(struct pf_state_key_cmp *key, u_int dir, int *more) 1417 { 1418 struct pf_keyhash *kh; 1419 struct pf_state_key *sk; 1420 struct pf_kstate *s, *ret = NULL; 1421 int idx, inout = 0; 1422 1423 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_SEARCH], 1); 1424 1425 kh = &V_pf_keyhash[pf_hashkey((struct pf_state_key *)key)]; 1426 1427 PF_HASHROW_LOCK(kh); 1428 LIST_FOREACH(sk, &kh->keys, entry) 1429 if (bcmp(sk, key, sizeof(struct pf_state_key_cmp)) == 0) 1430 break; 1431 if (sk == NULL) { 1432 PF_HASHROW_UNLOCK(kh); 1433 return (NULL); 1434 } 1435 switch (dir) { 1436 case PF_IN: 1437 idx = PF_SK_WIRE; 1438 break; 1439 case PF_OUT: 1440 idx = PF_SK_STACK; 1441 break; 1442 case PF_INOUT: 1443 idx = PF_SK_WIRE; 1444 inout = 1; 1445 break; 1446 default: 1447 panic("%s: dir %u", __func__, dir); 1448 } 1449 second_run: 1450 TAILQ_FOREACH(s, &sk->states[idx], key_list[idx]) { 1451 if (more == NULL) { 1452 PF_HASHROW_UNLOCK(kh); 1453 return (s); 1454 } 1455 1456 if (ret) 1457 (*more)++; 1458 else 1459 ret = s; 1460 } 1461 if (inout == 1) { 1462 inout = 0; 1463 idx = PF_SK_STACK; 1464 goto second_run; 1465 } 1466 PF_HASHROW_UNLOCK(kh); 1467 1468 return (ret); 1469 } 1470 1471 bool 1472 pf_find_state_all_exists(struct pf_state_key_cmp *key, u_int dir) 1473 { 1474 struct pf_kstate *s; 1475 1476 s = pf_find_state_all(key, dir, NULL); 1477 return (s != NULL); 1478 } 1479 1480 /* END state table stuff */ 1481 1482 static void 1483 pf_send(struct pf_send_entry *pfse) 1484 { 1485 1486 PF_SENDQ_LOCK(); 1487 STAILQ_INSERT_TAIL(&V_pf_sendqueue, pfse, pfse_next); 1488 PF_SENDQ_UNLOCK(); 1489 swi_sched(V_pf_swi_cookie, 0); 1490 } 1491 1492 static bool 1493 pf_isforlocal(struct mbuf *m, int af) 1494 { 1495 switch (af) { 1496 #ifdef INET 1497 case AF_INET: { 1498 struct rm_priotracker in_ifa_tracker; 1499 struct ip *ip; 1500 struct in_ifaddr *ia = NULL; 1501 1502 ip = mtod(m, struct ip *); 1503 IN_IFADDR_RLOCK(&in_ifa_tracker); 1504 LIST_FOREACH(ia, INADDR_HASH(ip->ip_dst.s_addr), ia_hash) { 1505 if (IA_SIN(ia)->sin_addr.s_addr == ip->ip_dst.s_addr) { 1506 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1507 return (true); 1508 } 1509 } 1510 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1511 break; 1512 } 1513 #endif 1514 #ifdef INET6 1515 case AF_INET6: { 1516 struct ip6_hdr *ip6; 1517 struct in6_ifaddr *ia; 1518 ip6 = mtod(m, struct ip6_hdr *); 1519 ia = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1520 if (ia == NULL) 1521 return (false); 1522 return (! (ia->ia6_flags & IN6_IFF_NOTREADY)); 1523 } 1524 #endif 1525 default: 1526 panic("Unsupported af %d", af); 1527 } 1528 1529 return (false); 1530 } 1531 1532 void 1533 pf_intr(void *v) 1534 { 1535 struct epoch_tracker et; 1536 struct pf_send_head queue; 1537 struct pf_send_entry *pfse, *next; 1538 1539 CURVNET_SET((struct vnet *)v); 1540 1541 PF_SENDQ_LOCK(); 1542 queue = V_pf_sendqueue; 1543 STAILQ_INIT(&V_pf_sendqueue); 1544 PF_SENDQ_UNLOCK(); 1545 1546 NET_EPOCH_ENTER(et); 1547 1548 STAILQ_FOREACH_SAFE(pfse, &queue, pfse_next, next) { 1549 switch (pfse->pfse_type) { 1550 #ifdef INET 1551 case PFSE_IP: { 1552 if (pf_isforlocal(pfse->pfse_m, AF_INET)) { 1553 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1554 pfse->pfse_m->m_pkthdr.csum_flags |= 1555 CSUM_IP_VALID | CSUM_IP_CHECKED; 1556 ip_input(pfse->pfse_m); 1557 } else { 1558 ip_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1559 NULL); 1560 } 1561 break; 1562 } 1563 case PFSE_ICMP: 1564 icmp_error(pfse->pfse_m, pfse->icmpopts.type, 1565 pfse->icmpopts.code, 0, pfse->icmpopts.mtu); 1566 break; 1567 #endif /* INET */ 1568 #ifdef INET6 1569 case PFSE_IP6: 1570 if (pf_isforlocal(pfse->pfse_m, AF_INET6)) { 1571 pfse->pfse_m->m_flags |= M_SKIP_FIREWALL; 1572 ip6_input(pfse->pfse_m); 1573 } else { 1574 ip6_output(pfse->pfse_m, NULL, NULL, 0, NULL, 1575 NULL, NULL); 1576 } 1577 break; 1578 case PFSE_ICMP6: 1579 icmp6_error(pfse->pfse_m, pfse->icmpopts.type, 1580 pfse->icmpopts.code, pfse->icmpopts.mtu); 1581 break; 1582 #endif /* INET6 */ 1583 default: 1584 panic("%s: unknown type", __func__); 1585 } 1586 free(pfse, M_PFTEMP); 1587 } 1588 NET_EPOCH_EXIT(et); 1589 CURVNET_RESTORE(); 1590 } 1591 1592 #define pf_purge_thread_period (hz / 10) 1593 1594 #ifdef PF_WANT_32_TO_64_COUNTER 1595 static void 1596 pf_status_counter_u64_periodic(void) 1597 { 1598 1599 PF_RULES_RASSERT(); 1600 1601 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 60)) != 0) { 1602 return; 1603 } 1604 1605 for (int i = 0; i < FCNT_MAX; i++) { 1606 pf_counter_u64_periodic(&V_pf_status.fcounters[i]); 1607 } 1608 } 1609 1610 static void 1611 pf_kif_counter_u64_periodic(void) 1612 { 1613 struct pfi_kkif *kif; 1614 size_t r, run; 1615 1616 PF_RULES_RASSERT(); 1617 1618 if (__predict_false(V_pf_allkifcount == 0)) { 1619 return; 1620 } 1621 1622 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 1623 return; 1624 } 1625 1626 run = V_pf_allkifcount / 10; 1627 if (run < 5) 1628 run = 5; 1629 1630 for (r = 0; r < run; r++) { 1631 kif = LIST_NEXT(V_pf_kifmarker, pfik_allkiflist); 1632 if (kif == NULL) { 1633 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 1634 LIST_INSERT_HEAD(&V_pf_allkiflist, V_pf_kifmarker, pfik_allkiflist); 1635 break; 1636 } 1637 1638 LIST_REMOVE(V_pf_kifmarker, pfik_allkiflist); 1639 LIST_INSERT_AFTER(kif, V_pf_kifmarker, pfik_allkiflist); 1640 1641 for (int i = 0; i < 2; i++) { 1642 for (int j = 0; j < 2; j++) { 1643 for (int k = 0; k < 2; k++) { 1644 pf_counter_u64_periodic(&kif->pfik_packets[i][j][k]); 1645 pf_counter_u64_periodic(&kif->pfik_bytes[i][j][k]); 1646 } 1647 } 1648 } 1649 } 1650 } 1651 1652 static void 1653 pf_rule_counter_u64_periodic(void) 1654 { 1655 struct pf_krule *rule; 1656 size_t r, run; 1657 1658 PF_RULES_RASSERT(); 1659 1660 if (__predict_false(V_pf_allrulecount == 0)) { 1661 return; 1662 } 1663 1664 if ((V_pf_counter_periodic_iter % (pf_purge_thread_period * 10 * 300)) != 0) { 1665 return; 1666 } 1667 1668 run = V_pf_allrulecount / 10; 1669 if (run < 5) 1670 run = 5; 1671 1672 for (r = 0; r < run; r++) { 1673 rule = LIST_NEXT(V_pf_rulemarker, allrulelist); 1674 if (rule == NULL) { 1675 LIST_REMOVE(V_pf_rulemarker, allrulelist); 1676 LIST_INSERT_HEAD(&V_pf_allrulelist, V_pf_rulemarker, allrulelist); 1677 break; 1678 } 1679 1680 LIST_REMOVE(V_pf_rulemarker, allrulelist); 1681 LIST_INSERT_AFTER(rule, V_pf_rulemarker, allrulelist); 1682 1683 pf_counter_u64_periodic(&rule->evaluations); 1684 for (int i = 0; i < 2; i++) { 1685 pf_counter_u64_periodic(&rule->packets[i]); 1686 pf_counter_u64_periodic(&rule->bytes[i]); 1687 } 1688 } 1689 } 1690 1691 static void 1692 pf_counter_u64_periodic_main(void) 1693 { 1694 PF_RULES_RLOCK_TRACKER; 1695 1696 V_pf_counter_periodic_iter++; 1697 1698 PF_RULES_RLOCK(); 1699 pf_counter_u64_critical_enter(); 1700 pf_status_counter_u64_periodic(); 1701 pf_kif_counter_u64_periodic(); 1702 pf_rule_counter_u64_periodic(); 1703 pf_counter_u64_critical_exit(); 1704 PF_RULES_RUNLOCK(); 1705 } 1706 #else 1707 #define pf_counter_u64_periodic_main() do { } while (0) 1708 #endif 1709 1710 void 1711 pf_purge_thread(void *unused __unused) 1712 { 1713 VNET_ITERATOR_DECL(vnet_iter); 1714 1715 sx_xlock(&pf_end_lock); 1716 while (pf_end_threads == 0) { 1717 sx_sleep(pf_purge_thread, &pf_end_lock, 0, "pftm", pf_purge_thread_period); 1718 1719 VNET_LIST_RLOCK(); 1720 VNET_FOREACH(vnet_iter) { 1721 CURVNET_SET(vnet_iter); 1722 1723 /* Wait until V_pf_default_rule is initialized. */ 1724 if (V_pf_vnet_active == 0) { 1725 CURVNET_RESTORE(); 1726 continue; 1727 } 1728 1729 pf_counter_u64_periodic_main(); 1730 1731 /* 1732 * Process 1/interval fraction of the state 1733 * table every run. 1734 */ 1735 V_pf_purge_idx = 1736 pf_purge_expired_states(V_pf_purge_idx, pf_hashmask / 1737 (V_pf_default_rule.timeout[PFTM_INTERVAL] * 10)); 1738 1739 /* 1740 * Purge other expired types every 1741 * PFTM_INTERVAL seconds. 1742 */ 1743 if (V_pf_purge_idx == 0) { 1744 /* 1745 * Order is important: 1746 * - states and src nodes reference rules 1747 * - states and rules reference kifs 1748 */ 1749 pf_purge_expired_fragments(); 1750 pf_purge_expired_src_nodes(); 1751 pf_purge_unlinked_rules(); 1752 pfi_kkif_purge(); 1753 } 1754 CURVNET_RESTORE(); 1755 } 1756 VNET_LIST_RUNLOCK(); 1757 } 1758 1759 pf_end_threads++; 1760 sx_xunlock(&pf_end_lock); 1761 kproc_exit(0); 1762 } 1763 1764 void 1765 pf_unload_vnet_purge(void) 1766 { 1767 1768 /* 1769 * To cleanse up all kifs and rules we need 1770 * two runs: first one clears reference flags, 1771 * then pf_purge_expired_states() doesn't 1772 * raise them, and then second run frees. 1773 */ 1774 pf_purge_unlinked_rules(); 1775 pfi_kkif_purge(); 1776 1777 /* 1778 * Now purge everything. 1779 */ 1780 pf_purge_expired_states(0, pf_hashmask); 1781 pf_purge_fragments(UINT_MAX); 1782 pf_purge_expired_src_nodes(); 1783 1784 /* 1785 * Now all kifs & rules should be unreferenced, 1786 * thus should be successfully freed. 1787 */ 1788 pf_purge_unlinked_rules(); 1789 pfi_kkif_purge(); 1790 } 1791 1792 u_int32_t 1793 pf_state_expires(const struct pf_kstate *state) 1794 { 1795 u_int32_t timeout; 1796 u_int32_t start; 1797 u_int32_t end; 1798 u_int32_t states; 1799 1800 /* handle all PFTM_* > PFTM_MAX here */ 1801 if (state->timeout == PFTM_PURGE) 1802 return (time_uptime); 1803 KASSERT(state->timeout != PFTM_UNLINKED, 1804 ("pf_state_expires: timeout == PFTM_UNLINKED")); 1805 KASSERT((state->timeout < PFTM_MAX), 1806 ("pf_state_expires: timeout > PFTM_MAX")); 1807 timeout = state->rule.ptr->timeout[state->timeout]; 1808 if (!timeout) 1809 timeout = V_pf_default_rule.timeout[state->timeout]; 1810 start = state->rule.ptr->timeout[PFTM_ADAPTIVE_START]; 1811 if (start && state->rule.ptr != &V_pf_default_rule) { 1812 end = state->rule.ptr->timeout[PFTM_ADAPTIVE_END]; 1813 states = counter_u64_fetch(state->rule.ptr->states_cur); 1814 } else { 1815 start = V_pf_default_rule.timeout[PFTM_ADAPTIVE_START]; 1816 end = V_pf_default_rule.timeout[PFTM_ADAPTIVE_END]; 1817 states = V_pf_status.states; 1818 } 1819 if (end && states > start && start < end) { 1820 if (states < end) { 1821 timeout = (u_int64_t)timeout * (end - states) / 1822 (end - start); 1823 return (state->expire + timeout); 1824 } 1825 else 1826 return (time_uptime); 1827 } 1828 return (state->expire + timeout); 1829 } 1830 1831 void 1832 pf_purge_expired_src_nodes() 1833 { 1834 struct pf_ksrc_node_list freelist; 1835 struct pf_srchash *sh; 1836 struct pf_ksrc_node *cur, *next; 1837 int i; 1838 1839 LIST_INIT(&freelist); 1840 for (i = 0, sh = V_pf_srchash; i <= pf_srchashmask; i++, sh++) { 1841 PF_HASHROW_LOCK(sh); 1842 LIST_FOREACH_SAFE(cur, &sh->nodes, entry, next) 1843 if (cur->states == 0 && cur->expire <= time_uptime) { 1844 pf_unlink_src_node(cur); 1845 LIST_INSERT_HEAD(&freelist, cur, entry); 1846 } else if (cur->rule.ptr != NULL) 1847 cur->rule.ptr->rule_ref |= PFRULE_REFS; 1848 PF_HASHROW_UNLOCK(sh); 1849 } 1850 1851 pf_free_src_nodes(&freelist); 1852 1853 V_pf_status.src_nodes = uma_zone_get_cur(V_pf_sources_z); 1854 } 1855 1856 static void 1857 pf_src_tree_remove_state(struct pf_kstate *s) 1858 { 1859 struct pf_ksrc_node *sn; 1860 struct pf_srchash *sh; 1861 uint32_t timeout; 1862 1863 timeout = s->rule.ptr->timeout[PFTM_SRC_NODE] ? 1864 s->rule.ptr->timeout[PFTM_SRC_NODE] : 1865 V_pf_default_rule.timeout[PFTM_SRC_NODE]; 1866 1867 if (s->src_node != NULL) { 1868 sn = s->src_node; 1869 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; 1870 PF_HASHROW_LOCK(sh); 1871 if (s->src.tcp_est) 1872 --sn->conn; 1873 if (--sn->states == 0) 1874 sn->expire = time_uptime + timeout; 1875 PF_HASHROW_UNLOCK(sh); 1876 } 1877 if (s->nat_src_node != s->src_node && s->nat_src_node != NULL) { 1878 sn = s->nat_src_node; 1879 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; 1880 PF_HASHROW_LOCK(sh); 1881 if (--sn->states == 0) 1882 sn->expire = time_uptime + timeout; 1883 PF_HASHROW_UNLOCK(sh); 1884 } 1885 s->src_node = s->nat_src_node = NULL; 1886 } 1887 1888 /* 1889 * Unlink and potentilly free a state. Function may be 1890 * called with ID hash row locked, but always returns 1891 * unlocked, since it needs to go through key hash locking. 1892 */ 1893 int 1894 pf_unlink_state(struct pf_kstate *s, u_int flags) 1895 { 1896 struct pf_idhash *ih = &V_pf_idhash[PF_IDHASH(s)]; 1897 1898 if ((flags & PF_ENTER_LOCKED) == 0) 1899 PF_HASHROW_LOCK(ih); 1900 else 1901 PF_HASHROW_ASSERT(ih); 1902 1903 if (s->timeout == PFTM_UNLINKED) { 1904 /* 1905 * State is being processed 1906 * by pf_unlink_state() in 1907 * an other thread. 1908 */ 1909 PF_HASHROW_UNLOCK(ih); 1910 return (0); /* XXXGL: undefined actually */ 1911 } 1912 1913 if (s->src.state == PF_TCPS_PROXY_DST) { 1914 /* XXX wire key the right one? */ 1915 pf_send_tcp(s->rule.ptr, s->key[PF_SK_WIRE]->af, 1916 &s->key[PF_SK_WIRE]->addr[1], 1917 &s->key[PF_SK_WIRE]->addr[0], 1918 s->key[PF_SK_WIRE]->port[1], 1919 s->key[PF_SK_WIRE]->port[0], 1920 s->src.seqhi, s->src.seqlo + 1, 1921 TH_RST|TH_ACK, 0, 0, 0, 1, s->tag); 1922 } 1923 1924 LIST_REMOVE(s, entry); 1925 pf_src_tree_remove_state(s); 1926 1927 if (V_pfsync_delete_state_ptr != NULL) 1928 V_pfsync_delete_state_ptr(s); 1929 1930 STATE_DEC_COUNTERS(s); 1931 1932 s->timeout = PFTM_UNLINKED; 1933 1934 PF_HASHROW_UNLOCK(ih); 1935 1936 pf_detach_state(s); 1937 /* pf_state_insert() initialises refs to 2 */ 1938 return (pf_release_staten(s, 2)); 1939 } 1940 1941 struct pf_kstate * 1942 pf_alloc_state(int flags) 1943 { 1944 1945 return (uma_zalloc(V_pf_state_z, flags | M_ZERO)); 1946 } 1947 1948 void 1949 pf_free_state(struct pf_kstate *cur) 1950 { 1951 1952 KASSERT(cur->refs == 0, ("%s: %p has refs", __func__, cur)); 1953 KASSERT(cur->timeout == PFTM_UNLINKED, ("%s: timeout %u", __func__, 1954 cur->timeout)); 1955 1956 pf_normalize_tcp_cleanup(cur); 1957 uma_zfree(V_pf_state_z, cur); 1958 pf_counter_u64_add(&V_pf_status.fcounters[FCNT_STATE_REMOVALS], 1); 1959 } 1960 1961 /* 1962 * Called only from pf_purge_thread(), thus serialized. 1963 */ 1964 static u_int 1965 pf_purge_expired_states(u_int i, int maxcheck) 1966 { 1967 struct pf_idhash *ih; 1968 struct pf_kstate *s; 1969 1970 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 1971 1972 /* 1973 * Go through hash and unlink states that expire now. 1974 */ 1975 while (maxcheck > 0) { 1976 ih = &V_pf_idhash[i]; 1977 1978 /* only take the lock if we expect to do work */ 1979 if (!LIST_EMPTY(&ih->states)) { 1980 relock: 1981 PF_HASHROW_LOCK(ih); 1982 LIST_FOREACH(s, &ih->states, entry) { 1983 if (pf_state_expires(s) <= time_uptime) { 1984 V_pf_status.states -= 1985 pf_unlink_state(s, PF_ENTER_LOCKED); 1986 goto relock; 1987 } 1988 s->rule.ptr->rule_ref |= PFRULE_REFS; 1989 if (s->nat_rule.ptr != NULL) 1990 s->nat_rule.ptr->rule_ref |= PFRULE_REFS; 1991 if (s->anchor.ptr != NULL) 1992 s->anchor.ptr->rule_ref |= PFRULE_REFS; 1993 s->kif->pfik_flags |= PFI_IFLAG_REFS; 1994 if (s->rt_kif) 1995 s->rt_kif->pfik_flags |= PFI_IFLAG_REFS; 1996 } 1997 PF_HASHROW_UNLOCK(ih); 1998 } 1999 2000 /* Return when we hit end of hash. */ 2001 if (++i > pf_hashmask) { 2002 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2003 return (0); 2004 } 2005 2006 maxcheck--; 2007 } 2008 2009 V_pf_status.states = uma_zone_get_cur(V_pf_state_z); 2010 2011 return (i); 2012 } 2013 2014 static void 2015 pf_purge_unlinked_rules() 2016 { 2017 struct pf_krulequeue tmpq; 2018 struct pf_krule *r, *r1; 2019 2020 /* 2021 * If we have overloading task pending, then we'd 2022 * better skip purging this time. There is a tiny 2023 * probability that overloading task references 2024 * an already unlinked rule. 2025 */ 2026 PF_OVERLOADQ_LOCK(); 2027 if (!SLIST_EMPTY(&V_pf_overloadqueue)) { 2028 PF_OVERLOADQ_UNLOCK(); 2029 return; 2030 } 2031 PF_OVERLOADQ_UNLOCK(); 2032 2033 /* 2034 * Do naive mark-and-sweep garbage collecting of old rules. 2035 * Reference flag is raised by pf_purge_expired_states() 2036 * and pf_purge_expired_src_nodes(). 2037 * 2038 * To avoid LOR between PF_UNLNKDRULES_LOCK/PF_RULES_WLOCK, 2039 * use a temporary queue. 2040 */ 2041 TAILQ_INIT(&tmpq); 2042 PF_UNLNKDRULES_LOCK(); 2043 TAILQ_FOREACH_SAFE(r, &V_pf_unlinked_rules, entries, r1) { 2044 if (!(r->rule_ref & PFRULE_REFS)) { 2045 TAILQ_REMOVE(&V_pf_unlinked_rules, r, entries); 2046 TAILQ_INSERT_TAIL(&tmpq, r, entries); 2047 } else 2048 r->rule_ref &= ~PFRULE_REFS; 2049 } 2050 PF_UNLNKDRULES_UNLOCK(); 2051 2052 if (!TAILQ_EMPTY(&tmpq)) { 2053 PF_RULES_WLOCK(); 2054 TAILQ_FOREACH_SAFE(r, &tmpq, entries, r1) { 2055 TAILQ_REMOVE(&tmpq, r, entries); 2056 pf_free_rule(r); 2057 } 2058 PF_RULES_WUNLOCK(); 2059 } 2060 } 2061 2062 void 2063 pf_print_host(struct pf_addr *addr, u_int16_t p, sa_family_t af) 2064 { 2065 switch (af) { 2066 #ifdef INET 2067 case AF_INET: { 2068 u_int32_t a = ntohl(addr->addr32[0]); 2069 printf("%u.%u.%u.%u", (a>>24)&255, (a>>16)&255, 2070 (a>>8)&255, a&255); 2071 if (p) { 2072 p = ntohs(p); 2073 printf(":%u", p); 2074 } 2075 break; 2076 } 2077 #endif /* INET */ 2078 #ifdef INET6 2079 case AF_INET6: { 2080 u_int16_t b; 2081 u_int8_t i, curstart, curend, maxstart, maxend; 2082 curstart = curend = maxstart = maxend = 255; 2083 for (i = 0; i < 8; i++) { 2084 if (!addr->addr16[i]) { 2085 if (curstart == 255) 2086 curstart = i; 2087 curend = i; 2088 } else { 2089 if ((curend - curstart) > 2090 (maxend - maxstart)) { 2091 maxstart = curstart; 2092 maxend = curend; 2093 } 2094 curstart = curend = 255; 2095 } 2096 } 2097 if ((curend - curstart) > 2098 (maxend - maxstart)) { 2099 maxstart = curstart; 2100 maxend = curend; 2101 } 2102 for (i = 0; i < 8; i++) { 2103 if (i >= maxstart && i <= maxend) { 2104 if (i == 0) 2105 printf(":"); 2106 if (i == maxend) 2107 printf(":"); 2108 } else { 2109 b = ntohs(addr->addr16[i]); 2110 printf("%x", b); 2111 if (i < 7) 2112 printf(":"); 2113 } 2114 } 2115 if (p) { 2116 p = ntohs(p); 2117 printf("[%u]", p); 2118 } 2119 break; 2120 } 2121 #endif /* INET6 */ 2122 } 2123 } 2124 2125 void 2126 pf_print_state(struct pf_kstate *s) 2127 { 2128 pf_print_state_parts(s, NULL, NULL); 2129 } 2130 2131 static void 2132 pf_print_state_parts(struct pf_kstate *s, 2133 struct pf_state_key *skwp, struct pf_state_key *sksp) 2134 { 2135 struct pf_state_key *skw, *sks; 2136 u_int8_t proto, dir; 2137 2138 /* Do our best to fill these, but they're skipped if NULL */ 2139 skw = skwp ? skwp : (s ? s->key[PF_SK_WIRE] : NULL); 2140 sks = sksp ? sksp : (s ? s->key[PF_SK_STACK] : NULL); 2141 proto = skw ? skw->proto : (sks ? sks->proto : 0); 2142 dir = s ? s->direction : 0; 2143 2144 switch (proto) { 2145 case IPPROTO_IPV4: 2146 printf("IPv4"); 2147 break; 2148 case IPPROTO_IPV6: 2149 printf("IPv6"); 2150 break; 2151 case IPPROTO_TCP: 2152 printf("TCP"); 2153 break; 2154 case IPPROTO_UDP: 2155 printf("UDP"); 2156 break; 2157 case IPPROTO_ICMP: 2158 printf("ICMP"); 2159 break; 2160 case IPPROTO_ICMPV6: 2161 printf("ICMPv6"); 2162 break; 2163 default: 2164 printf("%u", proto); 2165 break; 2166 } 2167 switch (dir) { 2168 case PF_IN: 2169 printf(" in"); 2170 break; 2171 case PF_OUT: 2172 printf(" out"); 2173 break; 2174 } 2175 if (skw) { 2176 printf(" wire: "); 2177 pf_print_host(&skw->addr[0], skw->port[0], skw->af); 2178 printf(" "); 2179 pf_print_host(&skw->addr[1], skw->port[1], skw->af); 2180 } 2181 if (sks) { 2182 printf(" stack: "); 2183 if (sks != skw) { 2184 pf_print_host(&sks->addr[0], sks->port[0], sks->af); 2185 printf(" "); 2186 pf_print_host(&sks->addr[1], sks->port[1], sks->af); 2187 } else 2188 printf("-"); 2189 } 2190 if (s) { 2191 if (proto == IPPROTO_TCP) { 2192 printf(" [lo=%u high=%u win=%u modulator=%u", 2193 s->src.seqlo, s->src.seqhi, 2194 s->src.max_win, s->src.seqdiff); 2195 if (s->src.wscale && s->dst.wscale) 2196 printf(" wscale=%u", 2197 s->src.wscale & PF_WSCALE_MASK); 2198 printf("]"); 2199 printf(" [lo=%u high=%u win=%u modulator=%u", 2200 s->dst.seqlo, s->dst.seqhi, 2201 s->dst.max_win, s->dst.seqdiff); 2202 if (s->src.wscale && s->dst.wscale) 2203 printf(" wscale=%u", 2204 s->dst.wscale & PF_WSCALE_MASK); 2205 printf("]"); 2206 } 2207 printf(" %u:%u", s->src.state, s->dst.state); 2208 } 2209 } 2210 2211 void 2212 pf_print_flags(u_int8_t f) 2213 { 2214 if (f) 2215 printf(" "); 2216 if (f & TH_FIN) 2217 printf("F"); 2218 if (f & TH_SYN) 2219 printf("S"); 2220 if (f & TH_RST) 2221 printf("R"); 2222 if (f & TH_PUSH) 2223 printf("P"); 2224 if (f & TH_ACK) 2225 printf("A"); 2226 if (f & TH_URG) 2227 printf("U"); 2228 if (f & TH_ECE) 2229 printf("E"); 2230 if (f & TH_CWR) 2231 printf("W"); 2232 } 2233 2234 #define PF_SET_SKIP_STEPS(i) \ 2235 do { \ 2236 while (head[i] != cur) { \ 2237 head[i]->skip[i].ptr = cur; \ 2238 head[i] = TAILQ_NEXT(head[i], entries); \ 2239 } \ 2240 } while (0) 2241 2242 void 2243 pf_calc_skip_steps(struct pf_krulequeue *rules) 2244 { 2245 struct pf_krule *cur, *prev, *head[PF_SKIP_COUNT]; 2246 int i; 2247 2248 cur = TAILQ_FIRST(rules); 2249 prev = cur; 2250 for (i = 0; i < PF_SKIP_COUNT; ++i) 2251 head[i] = cur; 2252 while (cur != NULL) { 2253 if (cur->kif != prev->kif || cur->ifnot != prev->ifnot) 2254 PF_SET_SKIP_STEPS(PF_SKIP_IFP); 2255 if (cur->direction != prev->direction) 2256 PF_SET_SKIP_STEPS(PF_SKIP_DIR); 2257 if (cur->af != prev->af) 2258 PF_SET_SKIP_STEPS(PF_SKIP_AF); 2259 if (cur->proto != prev->proto) 2260 PF_SET_SKIP_STEPS(PF_SKIP_PROTO); 2261 if (cur->src.neg != prev->src.neg || 2262 pf_addr_wrap_neq(&cur->src.addr, &prev->src.addr)) 2263 PF_SET_SKIP_STEPS(PF_SKIP_SRC_ADDR); 2264 if (cur->src.port[0] != prev->src.port[0] || 2265 cur->src.port[1] != prev->src.port[1] || 2266 cur->src.port_op != prev->src.port_op) 2267 PF_SET_SKIP_STEPS(PF_SKIP_SRC_PORT); 2268 if (cur->dst.neg != prev->dst.neg || 2269 pf_addr_wrap_neq(&cur->dst.addr, &prev->dst.addr)) 2270 PF_SET_SKIP_STEPS(PF_SKIP_DST_ADDR); 2271 if (cur->dst.port[0] != prev->dst.port[0] || 2272 cur->dst.port[1] != prev->dst.port[1] || 2273 cur->dst.port_op != prev->dst.port_op) 2274 PF_SET_SKIP_STEPS(PF_SKIP_DST_PORT); 2275 2276 prev = cur; 2277 cur = TAILQ_NEXT(cur, entries); 2278 } 2279 for (i = 0; i < PF_SKIP_COUNT; ++i) 2280 PF_SET_SKIP_STEPS(i); 2281 } 2282 2283 static int 2284 pf_addr_wrap_neq(struct pf_addr_wrap *aw1, struct pf_addr_wrap *aw2) 2285 { 2286 if (aw1->type != aw2->type) 2287 return (1); 2288 switch (aw1->type) { 2289 case PF_ADDR_ADDRMASK: 2290 case PF_ADDR_RANGE: 2291 if (PF_ANEQ(&aw1->v.a.addr, &aw2->v.a.addr, AF_INET6)) 2292 return (1); 2293 if (PF_ANEQ(&aw1->v.a.mask, &aw2->v.a.mask, AF_INET6)) 2294 return (1); 2295 return (0); 2296 case PF_ADDR_DYNIFTL: 2297 return (aw1->p.dyn->pfid_kt != aw2->p.dyn->pfid_kt); 2298 case PF_ADDR_NOROUTE: 2299 case PF_ADDR_URPFFAILED: 2300 return (0); 2301 case PF_ADDR_TABLE: 2302 return (aw1->p.tbl != aw2->p.tbl); 2303 default: 2304 printf("invalid address type: %d\n", aw1->type); 2305 return (1); 2306 } 2307 } 2308 2309 /** 2310 * Checksum updates are a little complicated because the checksum in the TCP/UDP 2311 * header isn't always a full checksum. In some cases (i.e. output) it's a 2312 * pseudo-header checksum, which is a partial checksum over src/dst IP 2313 * addresses, protocol number and length. 2314 * 2315 * That means we have the following cases: 2316 * * Input or forwarding: we don't have TSO, the checksum fields are full 2317 * checksums, we need to update the checksum whenever we change anything. 2318 * * Output (i.e. the checksum is a pseudo-header checksum): 2319 * x The field being updated is src/dst address or affects the length of 2320 * the packet. We need to update the pseudo-header checksum (note that this 2321 * checksum is not ones' complement). 2322 * x Some other field is being modified (e.g. src/dst port numbers): We 2323 * don't have to update anything. 2324 **/ 2325 u_int16_t 2326 pf_cksum_fixup(u_int16_t cksum, u_int16_t old, u_int16_t new, u_int8_t udp) 2327 { 2328 u_int32_t x; 2329 2330 x = cksum + old - new; 2331 x = (x + (x >> 16)) & 0xffff; 2332 2333 /* optimise: eliminate a branch when not udp */ 2334 if (udp && cksum == 0x0000) 2335 return cksum; 2336 if (udp && x == 0x0000) 2337 x = 0xffff; 2338 2339 return (u_int16_t)(x); 2340 } 2341 2342 static void 2343 pf_patch_8(struct mbuf *m, u_int16_t *cksum, u_int8_t *f, u_int8_t v, bool hi, 2344 u_int8_t udp) 2345 { 2346 u_int16_t old = htons(hi ? (*f << 8) : *f); 2347 u_int16_t new = htons(hi ? ( v << 8) : v); 2348 2349 if (*f == v) 2350 return; 2351 2352 *f = v; 2353 2354 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2355 return; 2356 2357 *cksum = pf_cksum_fixup(*cksum, old, new, udp); 2358 } 2359 2360 void 2361 pf_patch_16_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int16_t v, 2362 bool hi, u_int8_t udp) 2363 { 2364 u_int8_t *fb = (u_int8_t *)f; 2365 u_int8_t *vb = (u_int8_t *)&v; 2366 2367 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2368 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2369 } 2370 2371 void 2372 pf_patch_32_unaligned(struct mbuf *m, u_int16_t *cksum, void *f, u_int32_t v, 2373 bool hi, u_int8_t udp) 2374 { 2375 u_int8_t *fb = (u_int8_t *)f; 2376 u_int8_t *vb = (u_int8_t *)&v; 2377 2378 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2379 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2380 pf_patch_8(m, cksum, fb++, *vb++, hi, udp); 2381 pf_patch_8(m, cksum, fb++, *vb++, !hi, udp); 2382 } 2383 2384 u_int16_t 2385 pf_proto_cksum_fixup(struct mbuf *m, u_int16_t cksum, u_int16_t old, 2386 u_int16_t new, u_int8_t udp) 2387 { 2388 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2389 return (cksum); 2390 2391 return (pf_cksum_fixup(cksum, old, new, udp)); 2392 } 2393 2394 static void 2395 pf_change_ap(struct mbuf *m, struct pf_addr *a, u_int16_t *p, u_int16_t *ic, 2396 u_int16_t *pc, struct pf_addr *an, u_int16_t pn, u_int8_t u, 2397 sa_family_t af) 2398 { 2399 struct pf_addr ao; 2400 u_int16_t po = *p; 2401 2402 PF_ACPY(&ao, a, af); 2403 PF_ACPY(a, an, af); 2404 2405 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | CSUM_DELAY_DATA_IPV6)) 2406 *pc = ~*pc; 2407 2408 *p = pn; 2409 2410 switch (af) { 2411 #ifdef INET 2412 case AF_INET: 2413 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2414 ao.addr16[0], an->addr16[0], 0), 2415 ao.addr16[1], an->addr16[1], 0); 2416 *p = pn; 2417 2418 *pc = pf_cksum_fixup(pf_cksum_fixup(*pc, 2419 ao.addr16[0], an->addr16[0], u), 2420 ao.addr16[1], an->addr16[1], u); 2421 2422 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2423 break; 2424 #endif /* INET */ 2425 #ifdef INET6 2426 case AF_INET6: 2427 *pc = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2428 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2429 pf_cksum_fixup(pf_cksum_fixup(*pc, 2430 ao.addr16[0], an->addr16[0], u), 2431 ao.addr16[1], an->addr16[1], u), 2432 ao.addr16[2], an->addr16[2], u), 2433 ao.addr16[3], an->addr16[3], u), 2434 ao.addr16[4], an->addr16[4], u), 2435 ao.addr16[5], an->addr16[5], u), 2436 ao.addr16[6], an->addr16[6], u), 2437 ao.addr16[7], an->addr16[7], u); 2438 2439 *pc = pf_proto_cksum_fixup(m, *pc, po, pn, u); 2440 break; 2441 #endif /* INET6 */ 2442 } 2443 2444 if (m->m_pkthdr.csum_flags & (CSUM_DELAY_DATA | 2445 CSUM_DELAY_DATA_IPV6)) { 2446 *pc = ~*pc; 2447 if (! *pc) 2448 *pc = 0xffff; 2449 } 2450 } 2451 2452 /* Changes a u_int32_t. Uses a void * so there are no align restrictions */ 2453 void 2454 pf_change_a(void *a, u_int16_t *c, u_int32_t an, u_int8_t u) 2455 { 2456 u_int32_t ao; 2457 2458 memcpy(&ao, a, sizeof(ao)); 2459 memcpy(a, &an, sizeof(u_int32_t)); 2460 *c = pf_cksum_fixup(pf_cksum_fixup(*c, ao / 65536, an / 65536, u), 2461 ao % 65536, an % 65536, u); 2462 } 2463 2464 void 2465 pf_change_proto_a(struct mbuf *m, void *a, u_int16_t *c, u_int32_t an, u_int8_t udp) 2466 { 2467 u_int32_t ao; 2468 2469 memcpy(&ao, a, sizeof(ao)); 2470 memcpy(a, &an, sizeof(u_int32_t)); 2471 2472 *c = pf_proto_cksum_fixup(m, 2473 pf_proto_cksum_fixup(m, *c, ao / 65536, an / 65536, udp), 2474 ao % 65536, an % 65536, udp); 2475 } 2476 2477 #ifdef INET6 2478 static void 2479 pf_change_a6(struct pf_addr *a, u_int16_t *c, struct pf_addr *an, u_int8_t u) 2480 { 2481 struct pf_addr ao; 2482 2483 PF_ACPY(&ao, a, AF_INET6); 2484 PF_ACPY(a, an, AF_INET6); 2485 2486 *c = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2487 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2488 pf_cksum_fixup(pf_cksum_fixup(*c, 2489 ao.addr16[0], an->addr16[0], u), 2490 ao.addr16[1], an->addr16[1], u), 2491 ao.addr16[2], an->addr16[2], u), 2492 ao.addr16[3], an->addr16[3], u), 2493 ao.addr16[4], an->addr16[4], u), 2494 ao.addr16[5], an->addr16[5], u), 2495 ao.addr16[6], an->addr16[6], u), 2496 ao.addr16[7], an->addr16[7], u); 2497 } 2498 #endif /* INET6 */ 2499 2500 static void 2501 pf_change_icmp(struct pf_addr *ia, u_int16_t *ip, struct pf_addr *oa, 2502 struct pf_addr *na, u_int16_t np, u_int16_t *pc, u_int16_t *h2c, 2503 u_int16_t *ic, u_int16_t *hc, u_int8_t u, sa_family_t af) 2504 { 2505 struct pf_addr oia, ooa; 2506 2507 PF_ACPY(&oia, ia, af); 2508 if (oa) 2509 PF_ACPY(&ooa, oa, af); 2510 2511 /* Change inner protocol port, fix inner protocol checksum. */ 2512 if (ip != NULL) { 2513 u_int16_t oip = *ip; 2514 u_int32_t opc; 2515 2516 if (pc != NULL) 2517 opc = *pc; 2518 *ip = np; 2519 if (pc != NULL) 2520 *pc = pf_cksum_fixup(*pc, oip, *ip, u); 2521 *ic = pf_cksum_fixup(*ic, oip, *ip, 0); 2522 if (pc != NULL) 2523 *ic = pf_cksum_fixup(*ic, opc, *pc, 0); 2524 } 2525 /* Change inner ip address, fix inner ip and icmp checksums. */ 2526 PF_ACPY(ia, na, af); 2527 switch (af) { 2528 #ifdef INET 2529 case AF_INET: { 2530 u_int32_t oh2c = *h2c; 2531 2532 *h2c = pf_cksum_fixup(pf_cksum_fixup(*h2c, 2533 oia.addr16[0], ia->addr16[0], 0), 2534 oia.addr16[1], ia->addr16[1], 0); 2535 *ic = pf_cksum_fixup(pf_cksum_fixup(*ic, 2536 oia.addr16[0], ia->addr16[0], 0), 2537 oia.addr16[1], ia->addr16[1], 0); 2538 *ic = pf_cksum_fixup(*ic, oh2c, *h2c, 0); 2539 break; 2540 } 2541 #endif /* INET */ 2542 #ifdef INET6 2543 case AF_INET6: 2544 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2545 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2546 pf_cksum_fixup(pf_cksum_fixup(*ic, 2547 oia.addr16[0], ia->addr16[0], u), 2548 oia.addr16[1], ia->addr16[1], u), 2549 oia.addr16[2], ia->addr16[2], u), 2550 oia.addr16[3], ia->addr16[3], u), 2551 oia.addr16[4], ia->addr16[4], u), 2552 oia.addr16[5], ia->addr16[5], u), 2553 oia.addr16[6], ia->addr16[6], u), 2554 oia.addr16[7], ia->addr16[7], u); 2555 break; 2556 #endif /* INET6 */ 2557 } 2558 /* Outer ip address, fix outer ip or icmpv6 checksum, if necessary. */ 2559 if (oa) { 2560 PF_ACPY(oa, na, af); 2561 switch (af) { 2562 #ifdef INET 2563 case AF_INET: 2564 *hc = pf_cksum_fixup(pf_cksum_fixup(*hc, 2565 ooa.addr16[0], oa->addr16[0], 0), 2566 ooa.addr16[1], oa->addr16[1], 0); 2567 break; 2568 #endif /* INET */ 2569 #ifdef INET6 2570 case AF_INET6: 2571 *ic = pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2572 pf_cksum_fixup(pf_cksum_fixup(pf_cksum_fixup( 2573 pf_cksum_fixup(pf_cksum_fixup(*ic, 2574 ooa.addr16[0], oa->addr16[0], u), 2575 ooa.addr16[1], oa->addr16[1], u), 2576 ooa.addr16[2], oa->addr16[2], u), 2577 ooa.addr16[3], oa->addr16[3], u), 2578 ooa.addr16[4], oa->addr16[4], u), 2579 ooa.addr16[5], oa->addr16[5], u), 2580 ooa.addr16[6], oa->addr16[6], u), 2581 ooa.addr16[7], oa->addr16[7], u); 2582 break; 2583 #endif /* INET6 */ 2584 } 2585 } 2586 } 2587 2588 /* 2589 * Need to modulate the sequence numbers in the TCP SACK option 2590 * (credits to Krzysztof Pfaff for report and patch) 2591 */ 2592 static int 2593 pf_modulate_sack(struct mbuf *m, int off, struct pf_pdesc *pd, 2594 struct tcphdr *th, struct pf_state_peer *dst) 2595 { 2596 int hlen = (th->th_off << 2) - sizeof(*th), thoptlen = hlen; 2597 u_int8_t opts[TCP_MAXOLEN], *opt = opts; 2598 int copyback = 0, i, olen; 2599 struct sackblk sack; 2600 2601 #define TCPOLEN_SACKLEN (TCPOLEN_SACK + 2) 2602 if (hlen < TCPOLEN_SACKLEN || 2603 !pf_pull_hdr(m, off + sizeof(*th), opts, hlen, NULL, NULL, pd->af)) 2604 return 0; 2605 2606 while (hlen >= TCPOLEN_SACKLEN) { 2607 size_t startoff = opt - opts; 2608 olen = opt[1]; 2609 switch (*opt) { 2610 case TCPOPT_EOL: /* FALLTHROUGH */ 2611 case TCPOPT_NOP: 2612 opt++; 2613 hlen--; 2614 break; 2615 case TCPOPT_SACK: 2616 if (olen > hlen) 2617 olen = hlen; 2618 if (olen >= TCPOLEN_SACKLEN) { 2619 for (i = 2; i + TCPOLEN_SACK <= olen; 2620 i += TCPOLEN_SACK) { 2621 memcpy(&sack, &opt[i], sizeof(sack)); 2622 pf_patch_32_unaligned(m, 2623 &th->th_sum, &sack.start, 2624 htonl(ntohl(sack.start) - dst->seqdiff), 2625 PF_ALGNMNT(startoff), 2626 0); 2627 pf_patch_32_unaligned(m, &th->th_sum, 2628 &sack.end, 2629 htonl(ntohl(sack.end) - dst->seqdiff), 2630 PF_ALGNMNT(startoff), 2631 0); 2632 memcpy(&opt[i], &sack, sizeof(sack)); 2633 } 2634 copyback = 1; 2635 } 2636 /* FALLTHROUGH */ 2637 default: 2638 if (olen < 2) 2639 olen = 2; 2640 hlen -= olen; 2641 opt += olen; 2642 } 2643 } 2644 2645 if (copyback) 2646 m_copyback(m, off + sizeof(*th), thoptlen, (caddr_t)opts); 2647 return (copyback); 2648 } 2649 2650 struct mbuf * 2651 pf_build_tcp(const struct pf_krule *r, sa_family_t af, 2652 const struct pf_addr *saddr, const struct pf_addr *daddr, 2653 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2654 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 2655 u_int16_t rtag) 2656 { 2657 struct mbuf *m; 2658 int len, tlen; 2659 #ifdef INET 2660 struct ip *h = NULL; 2661 #endif /* INET */ 2662 #ifdef INET6 2663 struct ip6_hdr *h6 = NULL; 2664 #endif /* INET6 */ 2665 struct tcphdr *th; 2666 char *opt; 2667 struct pf_mtag *pf_mtag; 2668 2669 len = 0; 2670 th = NULL; 2671 2672 /* maximum segment size tcp option */ 2673 tlen = sizeof(struct tcphdr); 2674 if (mss) 2675 tlen += 4; 2676 2677 switch (af) { 2678 #ifdef INET 2679 case AF_INET: 2680 len = sizeof(struct ip) + tlen; 2681 break; 2682 #endif /* INET */ 2683 #ifdef INET6 2684 case AF_INET6: 2685 len = sizeof(struct ip6_hdr) + tlen; 2686 break; 2687 #endif /* INET6 */ 2688 default: 2689 panic("%s: unsupported af %d", __func__, af); 2690 } 2691 2692 m = m_gethdr(M_NOWAIT, MT_DATA); 2693 if (m == NULL) 2694 return (NULL); 2695 2696 #ifdef MAC 2697 mac_netinet_firewall_send(m); 2698 #endif 2699 if ((pf_mtag = pf_get_mtag(m)) == NULL) { 2700 m_freem(m); 2701 return (NULL); 2702 } 2703 if (tag) 2704 m->m_flags |= M_SKIP_FIREWALL; 2705 pf_mtag->tag = rtag; 2706 2707 if (r != NULL && r->rtableid >= 0) 2708 M_SETFIB(m, r->rtableid); 2709 2710 #ifdef ALTQ 2711 if (r != NULL && r->qid) { 2712 pf_mtag->qid = r->qid; 2713 2714 /* add hints for ecn */ 2715 pf_mtag->hdr = mtod(m, struct ip *); 2716 } 2717 #endif /* ALTQ */ 2718 m->m_data += max_linkhdr; 2719 m->m_pkthdr.len = m->m_len = len; 2720 /* The rest of the stack assumes a rcvif, so provide one. 2721 * This is a locally generated packet, so .. close enough. */ 2722 m->m_pkthdr.rcvif = V_loif; 2723 bzero(m->m_data, len); 2724 switch (af) { 2725 #ifdef INET 2726 case AF_INET: 2727 h = mtod(m, struct ip *); 2728 2729 /* IP header fields included in the TCP checksum */ 2730 h->ip_p = IPPROTO_TCP; 2731 h->ip_len = htons(tlen); 2732 h->ip_src.s_addr = saddr->v4.s_addr; 2733 h->ip_dst.s_addr = daddr->v4.s_addr; 2734 2735 th = (struct tcphdr *)((caddr_t)h + sizeof(struct ip)); 2736 break; 2737 #endif /* INET */ 2738 #ifdef INET6 2739 case AF_INET6: 2740 h6 = mtod(m, struct ip6_hdr *); 2741 2742 /* IP header fields included in the TCP checksum */ 2743 h6->ip6_nxt = IPPROTO_TCP; 2744 h6->ip6_plen = htons(tlen); 2745 memcpy(&h6->ip6_src, &saddr->v6, sizeof(struct in6_addr)); 2746 memcpy(&h6->ip6_dst, &daddr->v6, sizeof(struct in6_addr)); 2747 2748 th = (struct tcphdr *)((caddr_t)h6 + sizeof(struct ip6_hdr)); 2749 break; 2750 #endif /* INET6 */ 2751 } 2752 2753 /* TCP header */ 2754 th->th_sport = sport; 2755 th->th_dport = dport; 2756 th->th_seq = htonl(seq); 2757 th->th_ack = htonl(ack); 2758 th->th_off = tlen >> 2; 2759 th->th_flags = flags; 2760 th->th_win = htons(win); 2761 2762 if (mss) { 2763 opt = (char *)(th + 1); 2764 opt[0] = TCPOPT_MAXSEG; 2765 opt[1] = 4; 2766 HTONS(mss); 2767 bcopy((caddr_t)&mss, (caddr_t)(opt + 2), 2); 2768 } 2769 2770 switch (af) { 2771 #ifdef INET 2772 case AF_INET: 2773 /* TCP checksum */ 2774 th->th_sum = in_cksum(m, len); 2775 2776 /* Finish the IP header */ 2777 h->ip_v = 4; 2778 h->ip_hl = sizeof(*h) >> 2; 2779 h->ip_tos = IPTOS_LOWDELAY; 2780 h->ip_off = htons(V_path_mtu_discovery ? IP_DF : 0); 2781 h->ip_len = htons(len); 2782 h->ip_ttl = ttl ? ttl : V_ip_defttl; 2783 h->ip_sum = 0; 2784 break; 2785 #endif /* INET */ 2786 #ifdef INET6 2787 case AF_INET6: 2788 /* TCP checksum */ 2789 th->th_sum = in6_cksum(m, IPPROTO_TCP, 2790 sizeof(struct ip6_hdr), tlen); 2791 2792 h6->ip6_vfc |= IPV6_VERSION; 2793 h6->ip6_hlim = IPV6_DEFHLIM; 2794 break; 2795 #endif /* INET6 */ 2796 } 2797 2798 return (m); 2799 } 2800 2801 void 2802 pf_send_tcp(const struct pf_krule *r, sa_family_t af, 2803 const struct pf_addr *saddr, const struct pf_addr *daddr, 2804 u_int16_t sport, u_int16_t dport, u_int32_t seq, u_int32_t ack, 2805 u_int8_t flags, u_int16_t win, u_int16_t mss, u_int8_t ttl, int tag, 2806 u_int16_t rtag) 2807 { 2808 struct pf_send_entry *pfse; 2809 struct mbuf *m; 2810 2811 m = pf_build_tcp(r, af, saddr, daddr, sport, dport, seq, ack, flags, 2812 win, mss, ttl, tag, rtag); 2813 if (m == NULL) 2814 return; 2815 2816 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 2817 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 2818 if (pfse == NULL) { 2819 m_freem(m); 2820 return; 2821 } 2822 2823 switch (af) { 2824 #ifdef INET 2825 case AF_INET: 2826 pfse->pfse_type = PFSE_IP; 2827 break; 2828 #endif /* INET */ 2829 #ifdef INET6 2830 case AF_INET6: 2831 pfse->pfse_type = PFSE_IP6; 2832 break; 2833 #endif /* INET6 */ 2834 } 2835 2836 pfse->pfse_m = m; 2837 pf_send(pfse); 2838 } 2839 2840 static void 2841 pf_return(struct pf_krule *r, struct pf_krule *nr, struct pf_pdesc *pd, 2842 struct pf_state_key *sk, int off, struct mbuf *m, struct tcphdr *th, 2843 struct pfi_kkif *kif, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen, 2844 u_short *reason) 2845 { 2846 struct pf_addr * const saddr = pd->src; 2847 struct pf_addr * const daddr = pd->dst; 2848 sa_family_t af = pd->af; 2849 2850 /* undo NAT changes, if they have taken place */ 2851 if (nr != NULL) { 2852 PF_ACPY(saddr, &sk->addr[pd->sidx], af); 2853 PF_ACPY(daddr, &sk->addr[pd->didx], af); 2854 if (pd->sport) 2855 *pd->sport = sk->port[pd->sidx]; 2856 if (pd->dport) 2857 *pd->dport = sk->port[pd->didx]; 2858 if (pd->proto_sum) 2859 *pd->proto_sum = bproto_sum; 2860 if (pd->ip_sum) 2861 *pd->ip_sum = bip_sum; 2862 m_copyback(m, off, hdrlen, pd->hdr.any); 2863 } 2864 if (pd->proto == IPPROTO_TCP && 2865 ((r->rule_flag & PFRULE_RETURNRST) || 2866 (r->rule_flag & PFRULE_RETURN)) && 2867 !(th->th_flags & TH_RST)) { 2868 u_int32_t ack = ntohl(th->th_seq) + pd->p_len; 2869 int len = 0; 2870 #ifdef INET 2871 struct ip *h4; 2872 #endif 2873 #ifdef INET6 2874 struct ip6_hdr *h6; 2875 #endif 2876 2877 switch (af) { 2878 #ifdef INET 2879 case AF_INET: 2880 h4 = mtod(m, struct ip *); 2881 len = ntohs(h4->ip_len) - off; 2882 break; 2883 #endif 2884 #ifdef INET6 2885 case AF_INET6: 2886 h6 = mtod(m, struct ip6_hdr *); 2887 len = ntohs(h6->ip6_plen) - (off - sizeof(*h6)); 2888 break; 2889 #endif 2890 } 2891 2892 if (pf_check_proto_cksum(m, off, len, IPPROTO_TCP, af)) 2893 REASON_SET(reason, PFRES_PROTCKSUM); 2894 else { 2895 if (th->th_flags & TH_SYN) 2896 ack++; 2897 if (th->th_flags & TH_FIN) 2898 ack++; 2899 pf_send_tcp(r, af, pd->dst, 2900 pd->src, th->th_dport, th->th_sport, 2901 ntohl(th->th_ack), ack, TH_RST|TH_ACK, 0, 0, 2902 r->return_ttl, 1, 0); 2903 } 2904 } else if (pd->proto != IPPROTO_ICMP && af == AF_INET && 2905 r->return_icmp) 2906 pf_send_icmp(m, r->return_icmp >> 8, 2907 r->return_icmp & 255, af, r); 2908 else if (pd->proto != IPPROTO_ICMPV6 && af == AF_INET6 && 2909 r->return_icmp6) 2910 pf_send_icmp(m, r->return_icmp6 >> 8, 2911 r->return_icmp6 & 255, af, r); 2912 } 2913 2914 static int 2915 pf_match_ieee8021q_pcp(u_int8_t prio, struct mbuf *m) 2916 { 2917 struct m_tag *mtag; 2918 u_int8_t mpcp; 2919 2920 mtag = m_tag_locate(m, MTAG_8021Q, MTAG_8021Q_PCP_IN, NULL); 2921 if (mtag == NULL) 2922 return (0); 2923 2924 if (prio == PF_PRIO_ZERO) 2925 prio = 0; 2926 2927 mpcp = *(uint8_t *)(mtag + 1); 2928 2929 return (mpcp == prio); 2930 } 2931 2932 static void 2933 pf_send_icmp(struct mbuf *m, u_int8_t type, u_int8_t code, sa_family_t af, 2934 struct pf_krule *r) 2935 { 2936 struct pf_send_entry *pfse; 2937 struct mbuf *m0; 2938 struct pf_mtag *pf_mtag; 2939 2940 /* Allocate outgoing queue entry, mbuf and mbuf tag. */ 2941 pfse = malloc(sizeof(*pfse), M_PFTEMP, M_NOWAIT); 2942 if (pfse == NULL) 2943 return; 2944 2945 if ((m0 = m_copypacket(m, M_NOWAIT)) == NULL) { 2946 free(pfse, M_PFTEMP); 2947 return; 2948 } 2949 2950 if ((pf_mtag = pf_get_mtag(m0)) == NULL) { 2951 free(pfse, M_PFTEMP); 2952 return; 2953 } 2954 /* XXX: revisit */ 2955 m0->m_flags |= M_SKIP_FIREWALL; 2956 2957 if (r->rtableid >= 0) 2958 M_SETFIB(m0, r->rtableid); 2959 2960 #ifdef ALTQ 2961 if (r->qid) { 2962 pf_mtag->qid = r->qid; 2963 /* add hints for ecn */ 2964 pf_mtag->hdr = mtod(m0, struct ip *); 2965 } 2966 #endif /* ALTQ */ 2967 2968 switch (af) { 2969 #ifdef INET 2970 case AF_INET: 2971 pfse->pfse_type = PFSE_ICMP; 2972 break; 2973 #endif /* INET */ 2974 #ifdef INET6 2975 case AF_INET6: 2976 pfse->pfse_type = PFSE_ICMP6; 2977 break; 2978 #endif /* INET6 */ 2979 } 2980 pfse->pfse_m = m0; 2981 pfse->icmpopts.type = type; 2982 pfse->icmpopts.code = code; 2983 pf_send(pfse); 2984 } 2985 2986 /* 2987 * Return 1 if the addresses a and b match (with mask m), otherwise return 0. 2988 * If n is 0, they match if they are equal. If n is != 0, they match if they 2989 * are different. 2990 */ 2991 int 2992 pf_match_addr(u_int8_t n, struct pf_addr *a, struct pf_addr *m, 2993 struct pf_addr *b, sa_family_t af) 2994 { 2995 int match = 0; 2996 2997 switch (af) { 2998 #ifdef INET 2999 case AF_INET: 3000 if ((a->addr32[0] & m->addr32[0]) == 3001 (b->addr32[0] & m->addr32[0])) 3002 match++; 3003 break; 3004 #endif /* INET */ 3005 #ifdef INET6 3006 case AF_INET6: 3007 if (((a->addr32[0] & m->addr32[0]) == 3008 (b->addr32[0] & m->addr32[0])) && 3009 ((a->addr32[1] & m->addr32[1]) == 3010 (b->addr32[1] & m->addr32[1])) && 3011 ((a->addr32[2] & m->addr32[2]) == 3012 (b->addr32[2] & m->addr32[2])) && 3013 ((a->addr32[3] & m->addr32[3]) == 3014 (b->addr32[3] & m->addr32[3]))) 3015 match++; 3016 break; 3017 #endif /* INET6 */ 3018 } 3019 if (match) { 3020 if (n) 3021 return (0); 3022 else 3023 return (1); 3024 } else { 3025 if (n) 3026 return (1); 3027 else 3028 return (0); 3029 } 3030 } 3031 3032 /* 3033 * Return 1 if b <= a <= e, otherwise return 0. 3034 */ 3035 int 3036 pf_match_addr_range(struct pf_addr *b, struct pf_addr *e, 3037 struct pf_addr *a, sa_family_t af) 3038 { 3039 switch (af) { 3040 #ifdef INET 3041 case AF_INET: 3042 if ((ntohl(a->addr32[0]) < ntohl(b->addr32[0])) || 3043 (ntohl(a->addr32[0]) > ntohl(e->addr32[0]))) 3044 return (0); 3045 break; 3046 #endif /* INET */ 3047 #ifdef INET6 3048 case AF_INET6: { 3049 int i; 3050 3051 /* check a >= b */ 3052 for (i = 0; i < 4; ++i) 3053 if (ntohl(a->addr32[i]) > ntohl(b->addr32[i])) 3054 break; 3055 else if (ntohl(a->addr32[i]) < ntohl(b->addr32[i])) 3056 return (0); 3057 /* check a <= e */ 3058 for (i = 0; i < 4; ++i) 3059 if (ntohl(a->addr32[i]) < ntohl(e->addr32[i])) 3060 break; 3061 else if (ntohl(a->addr32[i]) > ntohl(e->addr32[i])) 3062 return (0); 3063 break; 3064 } 3065 #endif /* INET6 */ 3066 } 3067 return (1); 3068 } 3069 3070 static int 3071 pf_match(u_int8_t op, u_int32_t a1, u_int32_t a2, u_int32_t p) 3072 { 3073 switch (op) { 3074 case PF_OP_IRG: 3075 return ((p > a1) && (p < a2)); 3076 case PF_OP_XRG: 3077 return ((p < a1) || (p > a2)); 3078 case PF_OP_RRG: 3079 return ((p >= a1) && (p <= a2)); 3080 case PF_OP_EQ: 3081 return (p == a1); 3082 case PF_OP_NE: 3083 return (p != a1); 3084 case PF_OP_LT: 3085 return (p < a1); 3086 case PF_OP_LE: 3087 return (p <= a1); 3088 case PF_OP_GT: 3089 return (p > a1); 3090 case PF_OP_GE: 3091 return (p >= a1); 3092 } 3093 return (0); /* never reached */ 3094 } 3095 3096 int 3097 pf_match_port(u_int8_t op, u_int16_t a1, u_int16_t a2, u_int16_t p) 3098 { 3099 NTOHS(a1); 3100 NTOHS(a2); 3101 NTOHS(p); 3102 return (pf_match(op, a1, a2, p)); 3103 } 3104 3105 static int 3106 pf_match_uid(u_int8_t op, uid_t a1, uid_t a2, uid_t u) 3107 { 3108 if (u == UID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3109 return (0); 3110 return (pf_match(op, a1, a2, u)); 3111 } 3112 3113 static int 3114 pf_match_gid(u_int8_t op, gid_t a1, gid_t a2, gid_t g) 3115 { 3116 if (g == GID_MAX && op != PF_OP_EQ && op != PF_OP_NE) 3117 return (0); 3118 return (pf_match(op, a1, a2, g)); 3119 } 3120 3121 int 3122 pf_match_tag(struct mbuf *m, struct pf_krule *r, int *tag, int mtag) 3123 { 3124 if (*tag == -1) 3125 *tag = mtag; 3126 3127 return ((!r->match_tag_not && r->match_tag == *tag) || 3128 (r->match_tag_not && r->match_tag != *tag)); 3129 } 3130 3131 int 3132 pf_tag_packet(struct mbuf *m, struct pf_pdesc *pd, int tag) 3133 { 3134 3135 KASSERT(tag > 0, ("%s: tag %d", __func__, tag)); 3136 3137 if (pd->pf_mtag == NULL && ((pd->pf_mtag = pf_get_mtag(m)) == NULL)) 3138 return (ENOMEM); 3139 3140 pd->pf_mtag->tag = tag; 3141 3142 return (0); 3143 } 3144 3145 #define PF_ANCHOR_STACKSIZE 32 3146 struct pf_kanchor_stackframe { 3147 struct pf_kruleset *rs; 3148 struct pf_krule *r; /* XXX: + match bit */ 3149 struct pf_kanchor *child; 3150 }; 3151 3152 /* 3153 * XXX: We rely on malloc(9) returning pointer aligned addresses. 3154 */ 3155 #define PF_ANCHORSTACK_MATCH 0x00000001 3156 #define PF_ANCHORSTACK_MASK (PF_ANCHORSTACK_MATCH) 3157 3158 #define PF_ANCHOR_MATCH(f) ((uintptr_t)(f)->r & PF_ANCHORSTACK_MATCH) 3159 #define PF_ANCHOR_RULE(f) (struct pf_krule *) \ 3160 ((uintptr_t)(f)->r & ~PF_ANCHORSTACK_MASK) 3161 #define PF_ANCHOR_SET_MATCH(f) do { (f)->r = (void *) \ 3162 ((uintptr_t)(f)->r | PF_ANCHORSTACK_MATCH); \ 3163 } while (0) 3164 3165 void 3166 pf_step_into_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3167 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3168 int *match) 3169 { 3170 struct pf_kanchor_stackframe *f; 3171 3172 PF_RULES_RASSERT(); 3173 3174 if (match) 3175 *match = 0; 3176 if (*depth >= PF_ANCHOR_STACKSIZE) { 3177 printf("%s: anchor stack overflow on %s\n", 3178 __func__, (*r)->anchor->name); 3179 *r = TAILQ_NEXT(*r, entries); 3180 return; 3181 } else if (*depth == 0 && a != NULL) 3182 *a = *r; 3183 f = stack + (*depth)++; 3184 f->rs = *rs; 3185 f->r = *r; 3186 if ((*r)->anchor_wildcard) { 3187 struct pf_kanchor_node *parent = &(*r)->anchor->children; 3188 3189 if ((f->child = RB_MIN(pf_kanchor_node, parent)) == NULL) { 3190 *r = NULL; 3191 return; 3192 } 3193 *rs = &f->child->ruleset; 3194 } else { 3195 f->child = NULL; 3196 *rs = &(*r)->anchor->ruleset; 3197 } 3198 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3199 } 3200 3201 int 3202 pf_step_out_of_anchor(struct pf_kanchor_stackframe *stack, int *depth, 3203 struct pf_kruleset **rs, int n, struct pf_krule **r, struct pf_krule **a, 3204 int *match) 3205 { 3206 struct pf_kanchor_stackframe *f; 3207 struct pf_krule *fr; 3208 int quick = 0; 3209 3210 PF_RULES_RASSERT(); 3211 3212 do { 3213 if (*depth <= 0) 3214 break; 3215 f = stack + *depth - 1; 3216 fr = PF_ANCHOR_RULE(f); 3217 if (f->child != NULL) { 3218 struct pf_kanchor_node *parent; 3219 3220 /* 3221 * This block traverses through 3222 * a wildcard anchor. 3223 */ 3224 parent = &fr->anchor->children; 3225 if (match != NULL && *match) { 3226 /* 3227 * If any of "*" matched, then 3228 * "foo/ *" matched, mark frame 3229 * appropriately. 3230 */ 3231 PF_ANCHOR_SET_MATCH(f); 3232 *match = 0; 3233 } 3234 f->child = RB_NEXT(pf_kanchor_node, parent, f->child); 3235 if (f->child != NULL) { 3236 *rs = &f->child->ruleset; 3237 *r = TAILQ_FIRST((*rs)->rules[n].active.ptr); 3238 if (*r == NULL) 3239 continue; 3240 else 3241 break; 3242 } 3243 } 3244 (*depth)--; 3245 if (*depth == 0 && a != NULL) 3246 *a = NULL; 3247 *rs = f->rs; 3248 if (PF_ANCHOR_MATCH(f) || (match != NULL && *match)) 3249 quick = fr->quick; 3250 *r = TAILQ_NEXT(fr, entries); 3251 } while (*r == NULL); 3252 3253 return (quick); 3254 } 3255 3256 #ifdef INET6 3257 void 3258 pf_poolmask(struct pf_addr *naddr, struct pf_addr *raddr, 3259 struct pf_addr *rmask, struct pf_addr *saddr, sa_family_t af) 3260 { 3261 switch (af) { 3262 #ifdef INET 3263 case AF_INET: 3264 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3265 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3266 break; 3267 #endif /* INET */ 3268 case AF_INET6: 3269 naddr->addr32[0] = (raddr->addr32[0] & rmask->addr32[0]) | 3270 ((rmask->addr32[0] ^ 0xffffffff ) & saddr->addr32[0]); 3271 naddr->addr32[1] = (raddr->addr32[1] & rmask->addr32[1]) | 3272 ((rmask->addr32[1] ^ 0xffffffff ) & saddr->addr32[1]); 3273 naddr->addr32[2] = (raddr->addr32[2] & rmask->addr32[2]) | 3274 ((rmask->addr32[2] ^ 0xffffffff ) & saddr->addr32[2]); 3275 naddr->addr32[3] = (raddr->addr32[3] & rmask->addr32[3]) | 3276 ((rmask->addr32[3] ^ 0xffffffff ) & saddr->addr32[3]); 3277 break; 3278 } 3279 } 3280 3281 void 3282 pf_addr_inc(struct pf_addr *addr, sa_family_t af) 3283 { 3284 switch (af) { 3285 #ifdef INET 3286 case AF_INET: 3287 addr->addr32[0] = htonl(ntohl(addr->addr32[0]) + 1); 3288 break; 3289 #endif /* INET */ 3290 case AF_INET6: 3291 if (addr->addr32[3] == 0xffffffff) { 3292 addr->addr32[3] = 0; 3293 if (addr->addr32[2] == 0xffffffff) { 3294 addr->addr32[2] = 0; 3295 if (addr->addr32[1] == 0xffffffff) { 3296 addr->addr32[1] = 0; 3297 addr->addr32[0] = 3298 htonl(ntohl(addr->addr32[0]) + 1); 3299 } else 3300 addr->addr32[1] = 3301 htonl(ntohl(addr->addr32[1]) + 1); 3302 } else 3303 addr->addr32[2] = 3304 htonl(ntohl(addr->addr32[2]) + 1); 3305 } else 3306 addr->addr32[3] = 3307 htonl(ntohl(addr->addr32[3]) + 1); 3308 break; 3309 } 3310 } 3311 #endif /* INET6 */ 3312 3313 void 3314 pf_rule_to_actions(struct pf_krule *r, struct pf_rule_actions *a) 3315 { 3316 if (r->qid) 3317 a->qid = r->qid; 3318 if (r->pqid) 3319 a->pqid = r->pqid; 3320 if (r->dnpipe) 3321 a->dnpipe = r->dnpipe; 3322 if (r->dnrpipe) 3323 a->dnpipe = r->dnrpipe; 3324 if (r->free_flags & PFRULE_DN_IS_PIPE) 3325 a->flags |= PFRULE_DN_IS_PIPE; 3326 } 3327 3328 int 3329 pf_socket_lookup(int direction, struct pf_pdesc *pd, struct mbuf *m) 3330 { 3331 struct pf_addr *saddr, *daddr; 3332 u_int16_t sport, dport; 3333 struct inpcbinfo *pi; 3334 struct inpcb *inp; 3335 3336 pd->lookup.uid = UID_MAX; 3337 pd->lookup.gid = GID_MAX; 3338 3339 switch (pd->proto) { 3340 case IPPROTO_TCP: 3341 sport = pd->hdr.tcp.th_sport; 3342 dport = pd->hdr.tcp.th_dport; 3343 pi = &V_tcbinfo; 3344 break; 3345 case IPPROTO_UDP: 3346 sport = pd->hdr.udp.uh_sport; 3347 dport = pd->hdr.udp.uh_dport; 3348 pi = &V_udbinfo; 3349 break; 3350 default: 3351 return (-1); 3352 } 3353 if (direction == PF_IN) { 3354 saddr = pd->src; 3355 daddr = pd->dst; 3356 } else { 3357 u_int16_t p; 3358 3359 p = sport; 3360 sport = dport; 3361 dport = p; 3362 saddr = pd->dst; 3363 daddr = pd->src; 3364 } 3365 switch (pd->af) { 3366 #ifdef INET 3367 case AF_INET: 3368 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, daddr->v4, 3369 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3370 if (inp == NULL) { 3371 inp = in_pcblookup_mbuf(pi, saddr->v4, sport, 3372 daddr->v4, dport, INPLOOKUP_WILDCARD | 3373 INPLOOKUP_RLOCKPCB, NULL, m); 3374 if (inp == NULL) 3375 return (-1); 3376 } 3377 break; 3378 #endif /* INET */ 3379 #ifdef INET6 3380 case AF_INET6: 3381 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, &daddr->v6, 3382 dport, INPLOOKUP_RLOCKPCB, NULL, m); 3383 if (inp == NULL) { 3384 inp = in6_pcblookup_mbuf(pi, &saddr->v6, sport, 3385 &daddr->v6, dport, INPLOOKUP_WILDCARD | 3386 INPLOOKUP_RLOCKPCB, NULL, m); 3387 if (inp == NULL) 3388 return (-1); 3389 } 3390 break; 3391 #endif /* INET6 */ 3392 3393 default: 3394 return (-1); 3395 } 3396 INP_RLOCK_ASSERT(inp); 3397 pd->lookup.uid = inp->inp_cred->cr_uid; 3398 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 3399 INP_RUNLOCK(inp); 3400 3401 return (1); 3402 } 3403 3404 u_int8_t 3405 pf_get_wscale(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3406 { 3407 int hlen; 3408 u_int8_t hdr[60]; 3409 u_int8_t *opt, optlen; 3410 u_int8_t wscale = 0; 3411 3412 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3413 if (hlen <= sizeof(struct tcphdr)) 3414 return (0); 3415 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3416 return (0); 3417 opt = hdr + sizeof(struct tcphdr); 3418 hlen -= sizeof(struct tcphdr); 3419 while (hlen >= 3) { 3420 switch (*opt) { 3421 case TCPOPT_EOL: 3422 case TCPOPT_NOP: 3423 ++opt; 3424 --hlen; 3425 break; 3426 case TCPOPT_WINDOW: 3427 wscale = opt[2]; 3428 if (wscale > TCP_MAX_WINSHIFT) 3429 wscale = TCP_MAX_WINSHIFT; 3430 wscale |= PF_WSCALE_FLAG; 3431 /* FALLTHROUGH */ 3432 default: 3433 optlen = opt[1]; 3434 if (optlen < 2) 3435 optlen = 2; 3436 hlen -= optlen; 3437 opt += optlen; 3438 break; 3439 } 3440 } 3441 return (wscale); 3442 } 3443 3444 u_int16_t 3445 pf_get_mss(struct mbuf *m, int off, u_int16_t th_off, sa_family_t af) 3446 { 3447 int hlen; 3448 u_int8_t hdr[60]; 3449 u_int8_t *opt, optlen; 3450 u_int16_t mss = V_tcp_mssdflt; 3451 3452 hlen = th_off << 2; /* hlen <= sizeof(hdr) */ 3453 if (hlen <= sizeof(struct tcphdr)) 3454 return (0); 3455 if (!pf_pull_hdr(m, off, hdr, hlen, NULL, NULL, af)) 3456 return (0); 3457 opt = hdr + sizeof(struct tcphdr); 3458 hlen -= sizeof(struct tcphdr); 3459 while (hlen >= TCPOLEN_MAXSEG) { 3460 switch (*opt) { 3461 case TCPOPT_EOL: 3462 case TCPOPT_NOP: 3463 ++opt; 3464 --hlen; 3465 break; 3466 case TCPOPT_MAXSEG: 3467 bcopy((caddr_t)(opt + 2), (caddr_t)&mss, 2); 3468 NTOHS(mss); 3469 /* FALLTHROUGH */ 3470 default: 3471 optlen = opt[1]; 3472 if (optlen < 2) 3473 optlen = 2; 3474 hlen -= optlen; 3475 opt += optlen; 3476 break; 3477 } 3478 } 3479 return (mss); 3480 } 3481 3482 static u_int16_t 3483 pf_calc_mss(struct pf_addr *addr, sa_family_t af, int rtableid, u_int16_t offer) 3484 { 3485 struct nhop_object *nh; 3486 #ifdef INET6 3487 struct in6_addr dst6; 3488 uint32_t scopeid; 3489 #endif /* INET6 */ 3490 int hlen = 0; 3491 uint16_t mss = 0; 3492 3493 NET_EPOCH_ASSERT(); 3494 3495 switch (af) { 3496 #ifdef INET 3497 case AF_INET: 3498 hlen = sizeof(struct ip); 3499 nh = fib4_lookup(rtableid, addr->v4, 0, 0, 0); 3500 if (nh != NULL) 3501 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 3502 break; 3503 #endif /* INET */ 3504 #ifdef INET6 3505 case AF_INET6: 3506 hlen = sizeof(struct ip6_hdr); 3507 in6_splitscope(&addr->v6, &dst6, &scopeid); 3508 nh = fib6_lookup(rtableid, &dst6, scopeid, 0, 0); 3509 if (nh != NULL) 3510 mss = nh->nh_mtu - hlen - sizeof(struct tcphdr); 3511 break; 3512 #endif /* INET6 */ 3513 } 3514 3515 mss = max(V_tcp_mssdflt, mss); 3516 mss = min(mss, offer); 3517 mss = max(mss, 64); /* sanity - at least max opt space */ 3518 return (mss); 3519 } 3520 3521 static u_int32_t 3522 pf_tcp_iss(struct pf_pdesc *pd) 3523 { 3524 MD5_CTX ctx; 3525 u_int32_t digest[4]; 3526 3527 if (V_pf_tcp_secret_init == 0) { 3528 arc4random_buf(&V_pf_tcp_secret, sizeof(V_pf_tcp_secret)); 3529 MD5Init(&V_pf_tcp_secret_ctx); 3530 MD5Update(&V_pf_tcp_secret_ctx, V_pf_tcp_secret, 3531 sizeof(V_pf_tcp_secret)); 3532 V_pf_tcp_secret_init = 1; 3533 } 3534 3535 ctx = V_pf_tcp_secret_ctx; 3536 3537 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_sport, sizeof(u_short)); 3538 MD5Update(&ctx, (char *)&pd->hdr.tcp.th_dport, sizeof(u_short)); 3539 if (pd->af == AF_INET6) { 3540 MD5Update(&ctx, (char *)&pd->src->v6, sizeof(struct in6_addr)); 3541 MD5Update(&ctx, (char *)&pd->dst->v6, sizeof(struct in6_addr)); 3542 } else { 3543 MD5Update(&ctx, (char *)&pd->src->v4, sizeof(struct in_addr)); 3544 MD5Update(&ctx, (char *)&pd->dst->v4, sizeof(struct in_addr)); 3545 } 3546 MD5Final((u_char *)digest, &ctx); 3547 V_pf_tcp_iss_off += 4096; 3548 #define ISN_RANDOM_INCREMENT (4096 - 1) 3549 return (digest[0] + (arc4random() & ISN_RANDOM_INCREMENT) + 3550 V_pf_tcp_iss_off); 3551 #undef ISN_RANDOM_INCREMENT 3552 } 3553 3554 static int 3555 pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, int direction, 3556 struct pfi_kkif *kif, struct mbuf *m, int off, struct pf_pdesc *pd, 3557 struct pf_krule **am, struct pf_kruleset **rsm, struct inpcb *inp) 3558 { 3559 struct pf_krule *nr = NULL; 3560 struct pf_addr * const saddr = pd->src; 3561 struct pf_addr * const daddr = pd->dst; 3562 sa_family_t af = pd->af; 3563 struct pf_krule *r, *a = NULL; 3564 struct pf_kruleset *ruleset = NULL; 3565 struct pf_ksrc_node *nsn = NULL; 3566 struct tcphdr *th = &pd->hdr.tcp; 3567 struct pf_state_key *sk = NULL, *nk = NULL; 3568 u_short reason; 3569 int rewrite = 0, hdrlen = 0; 3570 int tag = -1, rtableid = -1; 3571 int asd = 0; 3572 int match = 0; 3573 int state_icmp = 0; 3574 u_int16_t sport = 0, dport = 0; 3575 u_int16_t bproto_sum = 0, bip_sum = 0; 3576 u_int8_t icmptype = 0, icmpcode = 0; 3577 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 3578 3579 PF_RULES_RASSERT(); 3580 3581 if (inp != NULL) { 3582 INP_LOCK_ASSERT(inp); 3583 pd->lookup.uid = inp->inp_cred->cr_uid; 3584 pd->lookup.gid = inp->inp_cred->cr_groups[0]; 3585 pd->lookup.done = 1; 3586 } 3587 3588 switch (pd->proto) { 3589 case IPPROTO_TCP: 3590 sport = th->th_sport; 3591 dport = th->th_dport; 3592 hdrlen = sizeof(*th); 3593 break; 3594 case IPPROTO_UDP: 3595 sport = pd->hdr.udp.uh_sport; 3596 dport = pd->hdr.udp.uh_dport; 3597 hdrlen = sizeof(pd->hdr.udp); 3598 break; 3599 #ifdef INET 3600 case IPPROTO_ICMP: 3601 if (pd->af != AF_INET) 3602 break; 3603 sport = dport = pd->hdr.icmp.icmp_id; 3604 hdrlen = sizeof(pd->hdr.icmp); 3605 icmptype = pd->hdr.icmp.icmp_type; 3606 icmpcode = pd->hdr.icmp.icmp_code; 3607 3608 if (icmptype == ICMP_UNREACH || 3609 icmptype == ICMP_SOURCEQUENCH || 3610 icmptype == ICMP_REDIRECT || 3611 icmptype == ICMP_TIMXCEED || 3612 icmptype == ICMP_PARAMPROB) 3613 state_icmp++; 3614 break; 3615 #endif /* INET */ 3616 #ifdef INET6 3617 case IPPROTO_ICMPV6: 3618 if (af != AF_INET6) 3619 break; 3620 sport = dport = pd->hdr.icmp6.icmp6_id; 3621 hdrlen = sizeof(pd->hdr.icmp6); 3622 icmptype = pd->hdr.icmp6.icmp6_type; 3623 icmpcode = pd->hdr.icmp6.icmp6_code; 3624 3625 if (icmptype == ICMP6_DST_UNREACH || 3626 icmptype == ICMP6_PACKET_TOO_BIG || 3627 icmptype == ICMP6_TIME_EXCEEDED || 3628 icmptype == ICMP6_PARAM_PROB) 3629 state_icmp++; 3630 break; 3631 #endif /* INET6 */ 3632 default: 3633 sport = dport = hdrlen = 0; 3634 break; 3635 } 3636 3637 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 3638 3639 /* check packet for BINAT/NAT/RDR */ 3640 if ((nr = pf_get_translation(pd, m, off, direction, kif, &nsn, &sk, 3641 &nk, saddr, daddr, sport, dport, anchor_stack)) != NULL) { 3642 KASSERT(sk != NULL, ("%s: null sk", __func__)); 3643 KASSERT(nk != NULL, ("%s: null nk", __func__)); 3644 3645 if (nr->log) { 3646 PFLOG_PACKET(kif, m, af, direction, PFRES_MATCH, nr, a, 3647 ruleset, pd, 1); 3648 } 3649 3650 if (pd->ip_sum) 3651 bip_sum = *pd->ip_sum; 3652 3653 switch (pd->proto) { 3654 case IPPROTO_TCP: 3655 bproto_sum = th->th_sum; 3656 pd->proto_sum = &th->th_sum; 3657 3658 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 3659 nk->port[pd->sidx] != sport) { 3660 pf_change_ap(m, saddr, &th->th_sport, pd->ip_sum, 3661 &th->th_sum, &nk->addr[pd->sidx], 3662 nk->port[pd->sidx], 0, af); 3663 pd->sport = &th->th_sport; 3664 sport = th->th_sport; 3665 } 3666 3667 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 3668 nk->port[pd->didx] != dport) { 3669 pf_change_ap(m, daddr, &th->th_dport, pd->ip_sum, 3670 &th->th_sum, &nk->addr[pd->didx], 3671 nk->port[pd->didx], 0, af); 3672 dport = th->th_dport; 3673 pd->dport = &th->th_dport; 3674 } 3675 rewrite++; 3676 break; 3677 case IPPROTO_UDP: 3678 bproto_sum = pd->hdr.udp.uh_sum; 3679 pd->proto_sum = &pd->hdr.udp.uh_sum; 3680 3681 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], af) || 3682 nk->port[pd->sidx] != sport) { 3683 pf_change_ap(m, saddr, &pd->hdr.udp.uh_sport, 3684 pd->ip_sum, &pd->hdr.udp.uh_sum, 3685 &nk->addr[pd->sidx], 3686 nk->port[pd->sidx], 1, af); 3687 sport = pd->hdr.udp.uh_sport; 3688 pd->sport = &pd->hdr.udp.uh_sport; 3689 } 3690 3691 if (PF_ANEQ(daddr, &nk->addr[pd->didx], af) || 3692 nk->port[pd->didx] != dport) { 3693 pf_change_ap(m, daddr, &pd->hdr.udp.uh_dport, 3694 pd->ip_sum, &pd->hdr.udp.uh_sum, 3695 &nk->addr[pd->didx], 3696 nk->port[pd->didx], 1, af); 3697 dport = pd->hdr.udp.uh_dport; 3698 pd->dport = &pd->hdr.udp.uh_dport; 3699 } 3700 rewrite++; 3701 break; 3702 #ifdef INET 3703 case IPPROTO_ICMP: 3704 nk->port[0] = nk->port[1]; 3705 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET)) 3706 pf_change_a(&saddr->v4.s_addr, pd->ip_sum, 3707 nk->addr[pd->sidx].v4.s_addr, 0); 3708 3709 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET)) 3710 pf_change_a(&daddr->v4.s_addr, pd->ip_sum, 3711 nk->addr[pd->didx].v4.s_addr, 0); 3712 3713 if (nk->port[1] != pd->hdr.icmp.icmp_id) { 3714 pd->hdr.icmp.icmp_cksum = pf_cksum_fixup( 3715 pd->hdr.icmp.icmp_cksum, sport, 3716 nk->port[1], 0); 3717 pd->hdr.icmp.icmp_id = nk->port[1]; 3718 pd->sport = &pd->hdr.icmp.icmp_id; 3719 } 3720 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 3721 break; 3722 #endif /* INET */ 3723 #ifdef INET6 3724 case IPPROTO_ICMPV6: 3725 nk->port[0] = nk->port[1]; 3726 if (PF_ANEQ(saddr, &nk->addr[pd->sidx], AF_INET6)) 3727 pf_change_a6(saddr, &pd->hdr.icmp6.icmp6_cksum, 3728 &nk->addr[pd->sidx], 0); 3729 3730 if (PF_ANEQ(daddr, &nk->addr[pd->didx], AF_INET6)) 3731 pf_change_a6(daddr, &pd->hdr.icmp6.icmp6_cksum, 3732 &nk->addr[pd->didx], 0); 3733 rewrite++; 3734 break; 3735 #endif /* INET */ 3736 default: 3737 switch (af) { 3738 #ifdef INET 3739 case AF_INET: 3740 if (PF_ANEQ(saddr, 3741 &nk->addr[pd->sidx], AF_INET)) 3742 pf_change_a(&saddr->v4.s_addr, 3743 pd->ip_sum, 3744 nk->addr[pd->sidx].v4.s_addr, 0); 3745 3746 if (PF_ANEQ(daddr, 3747 &nk->addr[pd->didx], AF_INET)) 3748 pf_change_a(&daddr->v4.s_addr, 3749 pd->ip_sum, 3750 nk->addr[pd->didx].v4.s_addr, 0); 3751 break; 3752 #endif /* INET */ 3753 #ifdef INET6 3754 case AF_INET6: 3755 if (PF_ANEQ(saddr, 3756 &nk->addr[pd->sidx], AF_INET6)) 3757 PF_ACPY(saddr, &nk->addr[pd->sidx], af); 3758 3759 if (PF_ANEQ(daddr, 3760 &nk->addr[pd->didx], AF_INET6)) 3761 PF_ACPY(daddr, &nk->addr[pd->didx], af); 3762 break; 3763 #endif /* INET */ 3764 } 3765 break; 3766 } 3767 if (nr->natpass) 3768 r = NULL; 3769 pd->nat_rule = nr; 3770 } 3771 3772 while (r != NULL) { 3773 pf_counter_u64_add(&r->evaluations, 1); 3774 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 3775 r = r->skip[PF_SKIP_IFP].ptr; 3776 else if (r->direction && r->direction != direction) 3777 r = r->skip[PF_SKIP_DIR].ptr; 3778 else if (r->af && r->af != af) 3779 r = r->skip[PF_SKIP_AF].ptr; 3780 else if (r->proto && r->proto != pd->proto) 3781 r = r->skip[PF_SKIP_PROTO].ptr; 3782 else if (PF_MISMATCHAW(&r->src.addr, saddr, af, 3783 r->src.neg, kif, M_GETFIB(m))) 3784 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 3785 /* tcp/udp only. port_op always 0 in other cases */ 3786 else if (r->src.port_op && !pf_match_port(r->src.port_op, 3787 r->src.port[0], r->src.port[1], sport)) 3788 r = r->skip[PF_SKIP_SRC_PORT].ptr; 3789 else if (PF_MISMATCHAW(&r->dst.addr, daddr, af, 3790 r->dst.neg, NULL, M_GETFIB(m))) 3791 r = r->skip[PF_SKIP_DST_ADDR].ptr; 3792 /* tcp/udp only. port_op always 0 in other cases */ 3793 else if (r->dst.port_op && !pf_match_port(r->dst.port_op, 3794 r->dst.port[0], r->dst.port[1], dport)) 3795 r = r->skip[PF_SKIP_DST_PORT].ptr; 3796 /* icmp only. type always 0 in other cases */ 3797 else if (r->type && r->type != icmptype + 1) 3798 r = TAILQ_NEXT(r, entries); 3799 /* icmp only. type always 0 in other cases */ 3800 else if (r->code && r->code != icmpcode + 1) 3801 r = TAILQ_NEXT(r, entries); 3802 else if (r->tos && !(r->tos == pd->tos)) 3803 r = TAILQ_NEXT(r, entries); 3804 else if (r->rule_flag & PFRULE_FRAGMENT) 3805 r = TAILQ_NEXT(r, entries); 3806 else if (pd->proto == IPPROTO_TCP && 3807 (r->flagset & th->th_flags) != r->flags) 3808 r = TAILQ_NEXT(r, entries); 3809 /* tcp/udp only. uid.op always 0 in other cases */ 3810 else if (r->uid.op && (pd->lookup.done || (pd->lookup.done = 3811 pf_socket_lookup(direction, pd, m), 1)) && 3812 !pf_match_uid(r->uid.op, r->uid.uid[0], r->uid.uid[1], 3813 pd->lookup.uid)) 3814 r = TAILQ_NEXT(r, entries); 3815 /* tcp/udp only. gid.op always 0 in other cases */ 3816 else if (r->gid.op && (pd->lookup.done || (pd->lookup.done = 3817 pf_socket_lookup(direction, pd, m), 1)) && 3818 !pf_match_gid(r->gid.op, r->gid.gid[0], r->gid.gid[1], 3819 pd->lookup.gid)) 3820 r = TAILQ_NEXT(r, entries); 3821 else if (r->prio && 3822 !pf_match_ieee8021q_pcp(r->prio, m)) 3823 r = TAILQ_NEXT(r, entries); 3824 else if (r->prob && 3825 r->prob <= arc4random()) 3826 r = TAILQ_NEXT(r, entries); 3827 else if (r->match_tag && !pf_match_tag(m, r, &tag, 3828 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 3829 r = TAILQ_NEXT(r, entries); 3830 else if (r->os_fingerprint != PF_OSFP_ANY && 3831 (pd->proto != IPPROTO_TCP || !pf_osfp_match( 3832 pf_osfp_fingerprint(pd, m, off, th), 3833 r->os_fingerprint))) 3834 r = TAILQ_NEXT(r, entries); 3835 else { 3836 if (r->tag) 3837 tag = r->tag; 3838 if (r->rtableid >= 0) 3839 rtableid = r->rtableid; 3840 if (r->anchor == NULL) { 3841 if (r->action == PF_MATCH) { 3842 pf_counter_u64_critical_enter(); 3843 pf_counter_u64_add_protected(&r->packets[direction == PF_OUT], 1); 3844 pf_counter_u64_add_protected(&r->bytes[direction == PF_OUT], pd->tot_len); 3845 pf_counter_u64_critical_exit(); 3846 pf_rule_to_actions(r, &pd->act); 3847 if (r->log) 3848 PFLOG_PACKET(kif, m, af, 3849 direction, PFRES_MATCH, r, 3850 a, ruleset, pd, 1); 3851 } else { 3852 match = 1; 3853 *rm = r; 3854 *am = a; 3855 *rsm = ruleset; 3856 } 3857 if ((*rm)->quick) 3858 break; 3859 r = TAILQ_NEXT(r, entries); 3860 } else 3861 pf_step_into_anchor(anchor_stack, &asd, 3862 &ruleset, PF_RULESET_FILTER, &r, &a, 3863 &match); 3864 } 3865 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 3866 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 3867 break; 3868 } 3869 r = *rm; 3870 a = *am; 3871 ruleset = *rsm; 3872 3873 REASON_SET(&reason, PFRES_MATCH); 3874 3875 /* apply actions for last matching pass/block rule */ 3876 pf_rule_to_actions(r, &pd->act); 3877 3878 if (r->log) { 3879 if (rewrite) 3880 m_copyback(m, off, hdrlen, pd->hdr.any); 3881 PFLOG_PACKET(kif, m, af, direction, reason, r, a, 3882 ruleset, pd, 1); 3883 } 3884 3885 if ((r->action == PF_DROP) && 3886 ((r->rule_flag & PFRULE_RETURNRST) || 3887 (r->rule_flag & PFRULE_RETURNICMP) || 3888 (r->rule_flag & PFRULE_RETURN))) { 3889 pf_return(r, nr, pd, sk, off, m, th, kif, bproto_sum, 3890 bip_sum, hdrlen, &reason); 3891 } 3892 3893 if (r->action == PF_DROP) 3894 goto cleanup; 3895 3896 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 3897 REASON_SET(&reason, PFRES_MEMORY); 3898 goto cleanup; 3899 } 3900 if (rtableid >= 0) 3901 M_SETFIB(m, rtableid); 3902 3903 if (!state_icmp && (r->keep_state || nr != NULL || 3904 (pd->flags & PFDESC_TCP_NORM))) { 3905 int action; 3906 action = pf_create_state(r, nr, a, pd, nsn, nk, sk, m, off, 3907 sport, dport, &rewrite, kif, sm, tag, bproto_sum, bip_sum, 3908 hdrlen); 3909 if (action != PF_PASS) { 3910 if (action == PF_DROP && 3911 (r->rule_flag & PFRULE_RETURN)) 3912 pf_return(r, nr, pd, sk, off, m, th, kif, 3913 bproto_sum, bip_sum, hdrlen, &reason); 3914 return (action); 3915 } 3916 } else { 3917 if (sk != NULL) 3918 uma_zfree(V_pf_state_key_z, sk); 3919 if (nk != NULL) 3920 uma_zfree(V_pf_state_key_z, nk); 3921 } 3922 3923 /* copy back packet headers if we performed NAT operations */ 3924 if (rewrite) 3925 m_copyback(m, off, hdrlen, pd->hdr.any); 3926 3927 if (*sm != NULL && !((*sm)->state_flags & PFSTATE_NOSYNC) && 3928 direction == PF_OUT && 3929 V_pfsync_defer_ptr != NULL && V_pfsync_defer_ptr(*sm, m)) 3930 /* 3931 * We want the state created, but we dont 3932 * want to send this in case a partner 3933 * firewall has to know about it to allow 3934 * replies through it. 3935 */ 3936 return (PF_DEFER); 3937 3938 return (PF_PASS); 3939 3940 cleanup: 3941 if (sk != NULL) 3942 uma_zfree(V_pf_state_key_z, sk); 3943 if (nk != NULL) 3944 uma_zfree(V_pf_state_key_z, nk); 3945 return (PF_DROP); 3946 } 3947 3948 static int 3949 pf_create_state(struct pf_krule *r, struct pf_krule *nr, struct pf_krule *a, 3950 struct pf_pdesc *pd, struct pf_ksrc_node *nsn, struct pf_state_key *nk, 3951 struct pf_state_key *sk, struct mbuf *m, int off, u_int16_t sport, 3952 u_int16_t dport, int *rewrite, struct pfi_kkif *kif, struct pf_kstate **sm, 3953 int tag, u_int16_t bproto_sum, u_int16_t bip_sum, int hdrlen) 3954 { 3955 struct pf_kstate *s = NULL; 3956 struct pf_ksrc_node *sn = NULL; 3957 struct tcphdr *th = &pd->hdr.tcp; 3958 u_int16_t mss = V_tcp_mssdflt; 3959 u_short reason; 3960 3961 /* check maximums */ 3962 if (r->max_states && 3963 (counter_u64_fetch(r->states_cur) >= r->max_states)) { 3964 counter_u64_add(V_pf_status.lcounters[LCNT_STATES], 1); 3965 REASON_SET(&reason, PFRES_MAXSTATES); 3966 goto csfailed; 3967 } 3968 /* src node for filter rule */ 3969 if ((r->rule_flag & PFRULE_SRCTRACK || 3970 r->rpool.opts & PF_POOL_STICKYADDR) && 3971 pf_insert_src_node(&sn, r, pd->src, pd->af) != 0) { 3972 REASON_SET(&reason, PFRES_SRCLIMIT); 3973 goto csfailed; 3974 } 3975 /* src node for translation rule */ 3976 if (nr != NULL && (nr->rpool.opts & PF_POOL_STICKYADDR) && 3977 pf_insert_src_node(&nsn, nr, &sk->addr[pd->sidx], pd->af)) { 3978 REASON_SET(&reason, PFRES_SRCLIMIT); 3979 goto csfailed; 3980 } 3981 s = pf_alloc_state(M_NOWAIT); 3982 if (s == NULL) { 3983 REASON_SET(&reason, PFRES_MEMORY); 3984 goto csfailed; 3985 } 3986 s->rule.ptr = r; 3987 s->nat_rule.ptr = nr; 3988 s->anchor.ptr = a; 3989 STATE_INC_COUNTERS(s); 3990 if (r->allow_opts) 3991 s->state_flags |= PFSTATE_ALLOWOPTS; 3992 if (r->rule_flag & PFRULE_STATESLOPPY) 3993 s->state_flags |= PFSTATE_SLOPPY; 3994 s->log = r->log & PF_LOG_ALL; 3995 s->sync_state = PFSYNC_S_NONE; 3996 s->qid = pd->act.qid; 3997 s->pqid = pd->act.pqid; 3998 s->dnpipe = pd->act.dnpipe; 3999 s->dnrpipe = pd->act.dnrpipe; 4000 s->state_flags |= pd->act.flags; 4001 if (nr != NULL) 4002 s->log |= nr->log & PF_LOG_ALL; 4003 switch (pd->proto) { 4004 case IPPROTO_TCP: 4005 s->src.seqlo = ntohl(th->th_seq); 4006 s->src.seqhi = s->src.seqlo + pd->p_len + 1; 4007 if ((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN && 4008 r->keep_state == PF_STATE_MODULATE) { 4009 /* Generate sequence number modulator */ 4010 if ((s->src.seqdiff = pf_tcp_iss(pd) - s->src.seqlo) == 4011 0) 4012 s->src.seqdiff = 1; 4013 pf_change_proto_a(m, &th->th_seq, &th->th_sum, 4014 htonl(s->src.seqlo + s->src.seqdiff), 0); 4015 *rewrite = 1; 4016 } else 4017 s->src.seqdiff = 0; 4018 if (th->th_flags & TH_SYN) { 4019 s->src.seqhi++; 4020 s->src.wscale = pf_get_wscale(m, off, 4021 th->th_off, pd->af); 4022 } 4023 s->src.max_win = MAX(ntohs(th->th_win), 1); 4024 if (s->src.wscale & PF_WSCALE_MASK) { 4025 /* Remove scale factor from initial window */ 4026 int win = s->src.max_win; 4027 win += 1 << (s->src.wscale & PF_WSCALE_MASK); 4028 s->src.max_win = (win - 1) >> 4029 (s->src.wscale & PF_WSCALE_MASK); 4030 } 4031 if (th->th_flags & TH_FIN) 4032 s->src.seqhi++; 4033 s->dst.seqhi = 1; 4034 s->dst.max_win = 1; 4035 pf_set_protostate(s, PF_PEER_SRC, TCPS_SYN_SENT); 4036 pf_set_protostate(s, PF_PEER_DST, TCPS_CLOSED); 4037 s->timeout = PFTM_TCP_FIRST_PACKET; 4038 break; 4039 case IPPROTO_UDP: 4040 pf_set_protostate(s, PF_PEER_SRC, PFUDPS_SINGLE); 4041 pf_set_protostate(s, PF_PEER_DST, PFUDPS_NO_TRAFFIC); 4042 s->timeout = PFTM_UDP_FIRST_PACKET; 4043 break; 4044 case IPPROTO_ICMP: 4045 #ifdef INET6 4046 case IPPROTO_ICMPV6: 4047 #endif 4048 s->timeout = PFTM_ICMP_FIRST_PACKET; 4049 break; 4050 default: 4051 pf_set_protostate(s, PF_PEER_SRC, PFOTHERS_SINGLE); 4052 pf_set_protostate(s, PF_PEER_DST, PFOTHERS_NO_TRAFFIC); 4053 s->timeout = PFTM_OTHER_FIRST_PACKET; 4054 } 4055 4056 if (r->rt) { 4057 if (pf_map_addr(pd->af, r, pd->src, &s->rt_addr, NULL, &sn)) { 4058 REASON_SET(&reason, PFRES_MAPFAILED); 4059 pf_src_tree_remove_state(s); 4060 s->timeout = PFTM_UNLINKED; 4061 STATE_DEC_COUNTERS(s); 4062 pf_free_state(s); 4063 goto csfailed; 4064 } 4065 s->rt_kif = r->rpool.cur->kif; 4066 } 4067 4068 s->creation = time_uptime; 4069 s->expire = time_uptime; 4070 4071 if (sn != NULL) 4072 s->src_node = sn; 4073 if (nsn != NULL) { 4074 /* XXX We only modify one side for now. */ 4075 PF_ACPY(&nsn->raddr, &nk->addr[1], pd->af); 4076 s->nat_src_node = nsn; 4077 } 4078 if (pd->proto == IPPROTO_TCP) { 4079 if ((pd->flags & PFDESC_TCP_NORM) && pf_normalize_tcp_init(m, 4080 off, pd, th, &s->src, &s->dst)) { 4081 REASON_SET(&reason, PFRES_MEMORY); 4082 pf_src_tree_remove_state(s); 4083 s->timeout = PFTM_UNLINKED; 4084 STATE_DEC_COUNTERS(s); 4085 pf_free_state(s); 4086 return (PF_DROP); 4087 } 4088 if ((pd->flags & PFDESC_TCP_NORM) && s->src.scrub && 4089 pf_normalize_tcp_stateful(m, off, pd, &reason, th, s, 4090 &s->src, &s->dst, rewrite)) { 4091 /* This really shouldn't happen!!! */ 4092 DPFPRINTF(PF_DEBUG_URGENT, 4093 ("pf_normalize_tcp_stateful failed on first " 4094 "pkt\n")); 4095 pf_src_tree_remove_state(s); 4096 s->timeout = PFTM_UNLINKED; 4097 STATE_DEC_COUNTERS(s); 4098 pf_free_state(s); 4099 return (PF_DROP); 4100 } 4101 } 4102 s->direction = pd->dir; 4103 4104 /* 4105 * sk/nk could already been setup by pf_get_translation(). 4106 */ 4107 if (nr == NULL) { 4108 KASSERT((sk == NULL && nk == NULL), ("%s: nr %p sk %p, nk %p", 4109 __func__, nr, sk, nk)); 4110 sk = pf_state_key_setup(pd, pd->src, pd->dst, sport, dport); 4111 if (sk == NULL) 4112 goto csfailed; 4113 nk = sk; 4114 } else 4115 KASSERT((sk != NULL && nk != NULL), ("%s: nr %p sk %p, nk %p", 4116 __func__, nr, sk, nk)); 4117 4118 /* Swap sk/nk for PF_OUT. */ 4119 if (pf_state_insert(BOUND_IFACE(r, kif), kif, 4120 (pd->dir == PF_IN) ? sk : nk, 4121 (pd->dir == PF_IN) ? nk : sk, s)) { 4122 REASON_SET(&reason, PFRES_STATEINS); 4123 pf_src_tree_remove_state(s); 4124 s->timeout = PFTM_UNLINKED; 4125 STATE_DEC_COUNTERS(s); 4126 pf_free_state(s); 4127 return (PF_DROP); 4128 } else 4129 *sm = s; 4130 4131 if (tag > 0) 4132 s->tag = tag; 4133 if (pd->proto == IPPROTO_TCP && (th->th_flags & (TH_SYN|TH_ACK)) == 4134 TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { 4135 pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); 4136 /* undo NAT changes, if they have taken place */ 4137 if (nr != NULL) { 4138 struct pf_state_key *skt = s->key[PF_SK_WIRE]; 4139 if (pd->dir == PF_OUT) 4140 skt = s->key[PF_SK_STACK]; 4141 PF_ACPY(pd->src, &skt->addr[pd->sidx], pd->af); 4142 PF_ACPY(pd->dst, &skt->addr[pd->didx], pd->af); 4143 if (pd->sport) 4144 *pd->sport = skt->port[pd->sidx]; 4145 if (pd->dport) 4146 *pd->dport = skt->port[pd->didx]; 4147 if (pd->proto_sum) 4148 *pd->proto_sum = bproto_sum; 4149 if (pd->ip_sum) 4150 *pd->ip_sum = bip_sum; 4151 m_copyback(m, off, hdrlen, pd->hdr.any); 4152 } 4153 s->src.seqhi = htonl(arc4random()); 4154 /* Find mss option */ 4155 int rtid = M_GETFIB(m); 4156 mss = pf_get_mss(m, off, th->th_off, pd->af); 4157 mss = pf_calc_mss(pd->src, pd->af, rtid, mss); 4158 mss = pf_calc_mss(pd->dst, pd->af, rtid, mss); 4159 s->src.mss = mss; 4160 pf_send_tcp(r, pd->af, pd->dst, pd->src, th->th_dport, 4161 th->th_sport, s->src.seqhi, ntohl(th->th_seq) + 1, 4162 TH_SYN|TH_ACK, 0, s->src.mss, 0, 1, 0); 4163 REASON_SET(&reason, PFRES_SYNPROXY); 4164 return (PF_SYNPROXY_DROP); 4165 } 4166 4167 return (PF_PASS); 4168 4169 csfailed: 4170 if (sk != NULL) 4171 uma_zfree(V_pf_state_key_z, sk); 4172 if (nk != NULL) 4173 uma_zfree(V_pf_state_key_z, nk); 4174 4175 if (sn != NULL) { 4176 struct pf_srchash *sh; 4177 4178 sh = &V_pf_srchash[pf_hashsrc(&sn->addr, sn->af)]; 4179 PF_HASHROW_LOCK(sh); 4180 if (--sn->states == 0 && sn->expire == 0) { 4181 pf_unlink_src_node(sn); 4182 uma_zfree(V_pf_sources_z, sn); 4183 counter_u64_add( 4184 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 4185 } 4186 PF_HASHROW_UNLOCK(sh); 4187 } 4188 4189 if (nsn != sn && nsn != NULL) { 4190 struct pf_srchash *sh; 4191 4192 sh = &V_pf_srchash[pf_hashsrc(&nsn->addr, nsn->af)]; 4193 PF_HASHROW_LOCK(sh); 4194 if (--nsn->states == 0 && nsn->expire == 0) { 4195 pf_unlink_src_node(nsn); 4196 uma_zfree(V_pf_sources_z, nsn); 4197 counter_u64_add( 4198 V_pf_status.scounters[SCNT_SRC_NODE_REMOVALS], 1); 4199 } 4200 PF_HASHROW_UNLOCK(sh); 4201 } 4202 4203 return (PF_DROP); 4204 } 4205 4206 static int 4207 pf_test_fragment(struct pf_krule **rm, int direction, struct pfi_kkif *kif, 4208 struct mbuf *m, void *h, struct pf_pdesc *pd, struct pf_krule **am, 4209 struct pf_kruleset **rsm) 4210 { 4211 struct pf_krule *r, *a = NULL; 4212 struct pf_kruleset *ruleset = NULL; 4213 sa_family_t af = pd->af; 4214 u_short reason; 4215 int tag = -1; 4216 int asd = 0; 4217 int match = 0; 4218 struct pf_kanchor_stackframe anchor_stack[PF_ANCHOR_STACKSIZE]; 4219 4220 PF_RULES_RASSERT(); 4221 4222 r = TAILQ_FIRST(pf_main_ruleset.rules[PF_RULESET_FILTER].active.ptr); 4223 while (r != NULL) { 4224 pf_counter_u64_add(&r->evaluations, 1); 4225 if (pfi_kkif_match(r->kif, kif) == r->ifnot) 4226 r = r->skip[PF_SKIP_IFP].ptr; 4227 else if (r->direction && r->direction != direction) 4228 r = r->skip[PF_SKIP_DIR].ptr; 4229 else if (r->af && r->af != af) 4230 r = r->skip[PF_SKIP_AF].ptr; 4231 else if (r->proto && r->proto != pd->proto) 4232 r = r->skip[PF_SKIP_PROTO].ptr; 4233 else if (PF_MISMATCHAW(&r->src.addr, pd->src, af, 4234 r->src.neg, kif, M_GETFIB(m))) 4235 r = r->skip[PF_SKIP_SRC_ADDR].ptr; 4236 else if (PF_MISMATCHAW(&r->dst.addr, pd->dst, af, 4237 r->dst.neg, NULL, M_GETFIB(m))) 4238 r = r->skip[PF_SKIP_DST_ADDR].ptr; 4239 else if (r->tos && !(r->tos == pd->tos)) 4240 r = TAILQ_NEXT(r, entries); 4241 else if (r->os_fingerprint != PF_OSFP_ANY) 4242 r = TAILQ_NEXT(r, entries); 4243 else if (pd->proto == IPPROTO_UDP && 4244 (r->src.port_op || r->dst.port_op)) 4245 r = TAILQ_NEXT(r, entries); 4246 else if (pd->proto == IPPROTO_TCP && 4247 (r->src.port_op || r->dst.port_op || r->flagset)) 4248 r = TAILQ_NEXT(r, entries); 4249 else if ((pd->proto == IPPROTO_ICMP || 4250 pd->proto == IPPROTO_ICMPV6) && 4251 (r->type || r->code)) 4252 r = TAILQ_NEXT(r, entries); 4253 else if (r->prio && 4254 !pf_match_ieee8021q_pcp(r->prio, m)) 4255 r = TAILQ_NEXT(r, entries); 4256 else if (r->prob && r->prob <= 4257 (arc4random() % (UINT_MAX - 1) + 1)) 4258 r = TAILQ_NEXT(r, entries); 4259 else if (r->match_tag && !pf_match_tag(m, r, &tag, 4260 pd->pf_mtag ? pd->pf_mtag->tag : 0)) 4261 r = TAILQ_NEXT(r, entries); 4262 else { 4263 if (r->anchor == NULL) { 4264 if (r->action == PF_MATCH) { 4265 pf_counter_u64_critical_enter(); 4266 pf_counter_u64_add_protected(&r->packets[direction == PF_OUT], 1); 4267 pf_counter_u64_add_protected(&r->bytes[direction == PF_OUT], pd->tot_len); 4268 pf_counter_u64_critical_exit(); 4269 pf_rule_to_actions(r, &pd->act); 4270 if (r->log) 4271 PFLOG_PACKET(kif, m, af, 4272 direction, PFRES_MATCH, r, 4273 a, ruleset, pd, 1); 4274 } else { 4275 match = 1; 4276 *rm = r; 4277 *am = a; 4278 *rsm = ruleset; 4279 } 4280 if ((*rm)->quick) 4281 break; 4282 r = TAILQ_NEXT(r, entries); 4283 } else 4284 pf_step_into_anchor(anchor_stack, &asd, 4285 &ruleset, PF_RULESET_FILTER, &r, &a, 4286 &match); 4287 } 4288 if (r == NULL && pf_step_out_of_anchor(anchor_stack, &asd, 4289 &ruleset, PF_RULESET_FILTER, &r, &a, &match)) 4290 break; 4291 } 4292 r = *rm; 4293 a = *am; 4294 ruleset = *rsm; 4295 4296 REASON_SET(&reason, PFRES_MATCH); 4297 4298 /* apply actions for last matching pass/block rule */ 4299 pf_rule_to_actions(r, &pd->act); 4300 4301 if (r->log) 4302 PFLOG_PACKET(kif, m, af, direction, reason, r, a, ruleset, pd, 4303 1); 4304 4305 if (r->action != PF_PASS) 4306 return (PF_DROP); 4307 4308 if (tag > 0 && pf_tag_packet(m, pd, tag)) { 4309 REASON_SET(&reason, PFRES_MEMORY); 4310 return (PF_DROP); 4311 } 4312 4313 return (PF_PASS); 4314 } 4315 4316 static int 4317 pf_tcp_track_full(struct pf_kstate **state, struct pfi_kkif *kif, 4318 struct mbuf *m, int off, struct pf_pdesc *pd, u_short *reason, 4319 int *copyback) 4320 { 4321 struct tcphdr *th = &pd->hdr.tcp; 4322 struct pf_state_peer *src, *dst; 4323 u_int16_t win = ntohs(th->th_win); 4324 u_int32_t ack, end, seq, orig_seq; 4325 u_int8_t sws, dws, psrc, pdst; 4326 int ackskew; 4327 4328 if (pd->dir == (*state)->direction) { 4329 src = &(*state)->src; 4330 dst = &(*state)->dst; 4331 psrc = PF_PEER_SRC; 4332 pdst = PF_PEER_DST; 4333 } else { 4334 src = &(*state)->dst; 4335 dst = &(*state)->src; 4336 psrc = PF_PEER_DST; 4337 pdst = PF_PEER_SRC; 4338 } 4339 4340 if (src->wscale && dst->wscale && !(th->th_flags & TH_SYN)) { 4341 sws = src->wscale & PF_WSCALE_MASK; 4342 dws = dst->wscale & PF_WSCALE_MASK; 4343 } else 4344 sws = dws = 0; 4345 4346 /* 4347 * Sequence tracking algorithm from Guido van Rooij's paper: 4348 * http://www.madison-gurkha.com/publications/tcp_filtering/ 4349 * tcp_filtering.ps 4350 */ 4351 4352 orig_seq = seq = ntohl(th->th_seq); 4353 if (src->seqlo == 0) { 4354 /* First packet from this end. Set its state */ 4355 4356 if ((pd->flags & PFDESC_TCP_NORM || dst->scrub) && 4357 src->scrub == NULL) { 4358 if (pf_normalize_tcp_init(m, off, pd, th, src, dst)) { 4359 REASON_SET(reason, PFRES_MEMORY); 4360 return (PF_DROP); 4361 } 4362 } 4363 4364 /* Deferred generation of sequence number modulator */ 4365 if (dst->seqdiff && !src->seqdiff) { 4366 /* use random iss for the TCP server */ 4367 while ((src->seqdiff = arc4random() - seq) == 0) 4368 ; 4369 ack = ntohl(th->th_ack) - dst->seqdiff; 4370 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 4371 src->seqdiff), 0); 4372 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 4373 *copyback = 1; 4374 } else { 4375 ack = ntohl(th->th_ack); 4376 } 4377 4378 end = seq + pd->p_len; 4379 if (th->th_flags & TH_SYN) { 4380 end++; 4381 if (dst->wscale & PF_WSCALE_FLAG) { 4382 src->wscale = pf_get_wscale(m, off, th->th_off, 4383 pd->af); 4384 if (src->wscale & PF_WSCALE_FLAG) { 4385 /* Remove scale factor from initial 4386 * window */ 4387 sws = src->wscale & PF_WSCALE_MASK; 4388 win = ((u_int32_t)win + (1 << sws) - 1) 4389 >> sws; 4390 dws = dst->wscale & PF_WSCALE_MASK; 4391 } else { 4392 /* fixup other window */ 4393 dst->max_win <<= dst->wscale & 4394 PF_WSCALE_MASK; 4395 /* in case of a retrans SYN|ACK */ 4396 dst->wscale = 0; 4397 } 4398 } 4399 } 4400 if (th->th_flags & TH_FIN) 4401 end++; 4402 4403 src->seqlo = seq; 4404 if (src->state < TCPS_SYN_SENT) 4405 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 4406 4407 /* 4408 * May need to slide the window (seqhi may have been set by 4409 * the crappy stack check or if we picked up the connection 4410 * after establishment) 4411 */ 4412 if (src->seqhi == 1 || 4413 SEQ_GEQ(end + MAX(1, dst->max_win << dws), src->seqhi)) 4414 src->seqhi = end + MAX(1, dst->max_win << dws); 4415 if (win > src->max_win) 4416 src->max_win = win; 4417 4418 } else { 4419 ack = ntohl(th->th_ack) - dst->seqdiff; 4420 if (src->seqdiff) { 4421 /* Modulate sequence numbers */ 4422 pf_change_proto_a(m, &th->th_seq, &th->th_sum, htonl(seq + 4423 src->seqdiff), 0); 4424 pf_change_proto_a(m, &th->th_ack, &th->th_sum, htonl(ack), 0); 4425 *copyback = 1; 4426 } 4427 end = seq + pd->p_len; 4428 if (th->th_flags & TH_SYN) 4429 end++; 4430 if (th->th_flags & TH_FIN) 4431 end++; 4432 } 4433 4434 if ((th->th_flags & TH_ACK) == 0) { 4435 /* Let it pass through the ack skew check */ 4436 ack = dst->seqlo; 4437 } else if ((ack == 0 && 4438 (th->th_flags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) || 4439 /* broken tcp stacks do not set ack */ 4440 (dst->state < TCPS_SYN_SENT)) { 4441 /* 4442 * Many stacks (ours included) will set the ACK number in an 4443 * FIN|ACK if the SYN times out -- no sequence to ACK. 4444 */ 4445 ack = dst->seqlo; 4446 } 4447 4448 if (seq == end) { 4449 /* Ease sequencing restrictions on no data packets */ 4450 seq = src->seqlo; 4451 end = seq; 4452 } 4453 4454 ackskew = dst->seqlo - ack; 4455 4456 /* 4457 * Need to demodulate the sequence numbers in any TCP SACK options 4458 * (Selective ACK). We could optionally validate the SACK values 4459 * against the current ACK window, either forwards or backwards, but 4460 * I'm not confident that SACK has been implemented properly 4461 * everywhere. It wouldn't surprise me if several stacks accidentally 4462 * SACK too far backwards of previously ACKed data. There really aren't 4463 * any security implications of bad SACKing unless the target stack 4464 * doesn't validate the option length correctly. Someone trying to 4465 * spoof into a TCP connection won't bother blindly sending SACK 4466 * options anyway. 4467 */ 4468 if (dst->seqdiff && (th->th_off << 2) > sizeof(struct tcphdr)) { 4469 if (pf_modulate_sack(m, off, pd, th, dst)) 4470 *copyback = 1; 4471 } 4472 4473 #define MAXACKWINDOW (0xffff + 1500) /* 1500 is an arbitrary fudge factor */ 4474 if (SEQ_GEQ(src->seqhi, end) && 4475 /* Last octet inside other's window space */ 4476 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) && 4477 /* Retrans: not more than one window back */ 4478 (ackskew >= -MAXACKWINDOW) && 4479 /* Acking not more than one reassembled fragment backwards */ 4480 (ackskew <= (MAXACKWINDOW << sws)) && 4481 /* Acking not more than one window forward */ 4482 ((th->th_flags & TH_RST) == 0 || orig_seq == src->seqlo || 4483 (orig_seq == src->seqlo + 1) || (orig_seq + 1 == src->seqlo) || 4484 (pd->flags & PFDESC_IP_REAS) == 0)) { 4485 /* Require an exact/+1 sequence match on resets when possible */ 4486 4487 if (dst->scrub || src->scrub) { 4488 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 4489 *state, src, dst, copyback)) 4490 return (PF_DROP); 4491 } 4492 4493 /* update max window */ 4494 if (src->max_win < win) 4495 src->max_win = win; 4496 /* synchronize sequencing */ 4497 if (SEQ_GT(end, src->seqlo)) 4498 src->seqlo = end; 4499 /* slide the window of what the other end can send */ 4500 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 4501 dst->seqhi = ack + MAX((win << sws), 1); 4502 4503 /* update states */ 4504 if (th->th_flags & TH_SYN) 4505 if (src->state < TCPS_SYN_SENT) 4506 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 4507 if (th->th_flags & TH_FIN) 4508 if (src->state < TCPS_CLOSING) 4509 pf_set_protostate(*state, psrc, TCPS_CLOSING); 4510 if (th->th_flags & TH_ACK) { 4511 if (dst->state == TCPS_SYN_SENT) { 4512 pf_set_protostate(*state, pdst, 4513 TCPS_ESTABLISHED); 4514 if (src->state == TCPS_ESTABLISHED && 4515 (*state)->src_node != NULL && 4516 pf_src_connlimit(state)) { 4517 REASON_SET(reason, PFRES_SRCLIMIT); 4518 return (PF_DROP); 4519 } 4520 } else if (dst->state == TCPS_CLOSING) 4521 pf_set_protostate(*state, pdst, 4522 TCPS_FIN_WAIT_2); 4523 } 4524 if (th->th_flags & TH_RST) 4525 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 4526 4527 /* update expire time */ 4528 (*state)->expire = time_uptime; 4529 if (src->state >= TCPS_FIN_WAIT_2 && 4530 dst->state >= TCPS_FIN_WAIT_2) 4531 (*state)->timeout = PFTM_TCP_CLOSED; 4532 else if (src->state >= TCPS_CLOSING && 4533 dst->state >= TCPS_CLOSING) 4534 (*state)->timeout = PFTM_TCP_FIN_WAIT; 4535 else if (src->state < TCPS_ESTABLISHED || 4536 dst->state < TCPS_ESTABLISHED) 4537 (*state)->timeout = PFTM_TCP_OPENING; 4538 else if (src->state >= TCPS_CLOSING || 4539 dst->state >= TCPS_CLOSING) 4540 (*state)->timeout = PFTM_TCP_CLOSING; 4541 else 4542 (*state)->timeout = PFTM_TCP_ESTABLISHED; 4543 4544 /* Fall through to PASS packet */ 4545 4546 } else if ((dst->state < TCPS_SYN_SENT || 4547 dst->state >= TCPS_FIN_WAIT_2 || 4548 src->state >= TCPS_FIN_WAIT_2) && 4549 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) && 4550 /* Within a window forward of the originating packet */ 4551 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW)) { 4552 /* Within a window backward of the originating packet */ 4553 4554 /* 4555 * This currently handles three situations: 4556 * 1) Stupid stacks will shotgun SYNs before their peer 4557 * replies. 4558 * 2) When PF catches an already established stream (the 4559 * firewall rebooted, the state table was flushed, routes 4560 * changed...) 4561 * 3) Packets get funky immediately after the connection 4562 * closes (this should catch Solaris spurious ACK|FINs 4563 * that web servers like to spew after a close) 4564 * 4565 * This must be a little more careful than the above code 4566 * since packet floods will also be caught here. We don't 4567 * update the TTL here to mitigate the damage of a packet 4568 * flood and so the same code can handle awkward establishment 4569 * and a loosened connection close. 4570 * In the establishment case, a correct peer response will 4571 * validate the connection, go through the normal state code 4572 * and keep updating the state TTL. 4573 */ 4574 4575 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4576 printf("pf: loose state match: "); 4577 pf_print_state(*state); 4578 pf_print_flags(th->th_flags); 4579 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 4580 "pkts=%llu:%llu dir=%s,%s\n", seq, orig_seq, ack, 4581 pd->p_len, ackskew, (unsigned long long)(*state)->packets[0], 4582 (unsigned long long)(*state)->packets[1], 4583 pd->dir == PF_IN ? "in" : "out", 4584 pd->dir == (*state)->direction ? "fwd" : "rev"); 4585 } 4586 4587 if (dst->scrub || src->scrub) { 4588 if (pf_normalize_tcp_stateful(m, off, pd, reason, th, 4589 *state, src, dst, copyback)) 4590 return (PF_DROP); 4591 } 4592 4593 /* update max window */ 4594 if (src->max_win < win) 4595 src->max_win = win; 4596 /* synchronize sequencing */ 4597 if (SEQ_GT(end, src->seqlo)) 4598 src->seqlo = end; 4599 /* slide the window of what the other end can send */ 4600 if (SEQ_GEQ(ack + (win << sws), dst->seqhi)) 4601 dst->seqhi = ack + MAX((win << sws), 1); 4602 4603 /* 4604 * Cannot set dst->seqhi here since this could be a shotgunned 4605 * SYN and not an already established connection. 4606 */ 4607 4608 if (th->th_flags & TH_FIN) 4609 if (src->state < TCPS_CLOSING) 4610 pf_set_protostate(*state, psrc, TCPS_CLOSING); 4611 if (th->th_flags & TH_RST) 4612 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 4613 4614 /* Fall through to PASS packet */ 4615 4616 } else { 4617 if ((*state)->dst.state == TCPS_SYN_SENT && 4618 (*state)->src.state == TCPS_SYN_SENT) { 4619 /* Send RST for state mismatches during handshake */ 4620 if (!(th->th_flags & TH_RST)) 4621 pf_send_tcp((*state)->rule.ptr, pd->af, 4622 pd->dst, pd->src, th->th_dport, 4623 th->th_sport, ntohl(th->th_ack), 0, 4624 TH_RST, 0, 0, 4625 (*state)->rule.ptr->return_ttl, 1, 0); 4626 src->seqlo = 0; 4627 src->seqhi = 1; 4628 src->max_win = 1; 4629 } else if (V_pf_status.debug >= PF_DEBUG_MISC) { 4630 printf("pf: BAD state: "); 4631 pf_print_state(*state); 4632 pf_print_flags(th->th_flags); 4633 printf(" seq=%u (%u) ack=%u len=%u ackskew=%d " 4634 "pkts=%llu:%llu dir=%s,%s\n", 4635 seq, orig_seq, ack, pd->p_len, ackskew, 4636 (unsigned long long)(*state)->packets[0], 4637 (unsigned long long)(*state)->packets[1], 4638 pd->dir == PF_IN ? "in" : "out", 4639 pd->dir == (*state)->direction ? "fwd" : "rev"); 4640 printf("pf: State failure on: %c %c %c %c | %c %c\n", 4641 SEQ_GEQ(src->seqhi, end) ? ' ' : '1', 4642 SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)) ? 4643 ' ': '2', 4644 (ackskew >= -MAXACKWINDOW) ? ' ' : '3', 4645 (ackskew <= (MAXACKWINDOW << sws)) ? ' ' : '4', 4646 SEQ_GEQ(src->seqhi + MAXACKWINDOW, end) ?' ' :'5', 4647 SEQ_GEQ(seq, src->seqlo - MAXACKWINDOW) ?' ' :'6'); 4648 } 4649 REASON_SET(reason, PFRES_BADSTATE); 4650 return (PF_DROP); 4651 } 4652 4653 return (PF_PASS); 4654 } 4655 4656 static int 4657 pf_tcp_track_sloppy(struct pf_kstate **state, struct pf_pdesc *pd, u_short *reason) 4658 { 4659 struct tcphdr *th = &pd->hdr.tcp; 4660 struct pf_state_peer *src, *dst; 4661 u_int8_t psrc, pdst; 4662 4663 if (pd->dir == (*state)->direction) { 4664 src = &(*state)->src; 4665 dst = &(*state)->dst; 4666 psrc = PF_PEER_SRC; 4667 pdst = PF_PEER_DST; 4668 } else { 4669 src = &(*state)->dst; 4670 dst = &(*state)->src; 4671 psrc = PF_PEER_DST; 4672 pdst = PF_PEER_SRC; 4673 } 4674 4675 if (th->th_flags & TH_SYN) 4676 if (src->state < TCPS_SYN_SENT) 4677 pf_set_protostate(*state, psrc, TCPS_SYN_SENT); 4678 if (th->th_flags & TH_FIN) 4679 if (src->state < TCPS_CLOSING) 4680 pf_set_protostate(*state, psrc, TCPS_CLOSING); 4681 if (th->th_flags & TH_ACK) { 4682 if (dst->state == TCPS_SYN_SENT) { 4683 pf_set_protostate(*state, pdst, TCPS_ESTABLISHED); 4684 if (src->state == TCPS_ESTABLISHED && 4685 (*state)->src_node != NULL && 4686 pf_src_connlimit(state)) { 4687 REASON_SET(reason, PFRES_SRCLIMIT); 4688 return (PF_DROP); 4689 } 4690 } else if (dst->state == TCPS_CLOSING) { 4691 pf_set_protostate(*state, pdst, TCPS_FIN_WAIT_2); 4692 } else if (src->state == TCPS_SYN_SENT && 4693 dst->state < TCPS_SYN_SENT) { 4694 /* 4695 * Handle a special sloppy case where we only see one 4696 * half of the connection. If there is a ACK after 4697 * the initial SYN without ever seeing a packet from 4698 * the destination, set the connection to established. 4699 */ 4700 pf_set_protostate(*state, PF_PEER_BOTH, 4701 TCPS_ESTABLISHED); 4702 dst->state = src->state = TCPS_ESTABLISHED; 4703 if ((*state)->src_node != NULL && 4704 pf_src_connlimit(state)) { 4705 REASON_SET(reason, PFRES_SRCLIMIT); 4706 return (PF_DROP); 4707 } 4708 } else if (src->state == TCPS_CLOSING && 4709 dst->state == TCPS_ESTABLISHED && 4710 dst->seqlo == 0) { 4711 /* 4712 * Handle the closing of half connections where we 4713 * don't see the full bidirectional FIN/ACK+ACK 4714 * handshake. 4715 */ 4716 pf_set_protostate(*state, pdst, TCPS_CLOSING); 4717 } 4718 } 4719 if (th->th_flags & TH_RST) 4720 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_TIME_WAIT); 4721 4722 /* update expire time */ 4723 (*state)->expire = time_uptime; 4724 if (src->state >= TCPS_FIN_WAIT_2 && 4725 dst->state >= TCPS_FIN_WAIT_2) 4726 (*state)->timeout = PFTM_TCP_CLOSED; 4727 else if (src->state >= TCPS_CLOSING && 4728 dst->state >= TCPS_CLOSING) 4729 (*state)->timeout = PFTM_TCP_FIN_WAIT; 4730 else if (src->state < TCPS_ESTABLISHED || 4731 dst->state < TCPS_ESTABLISHED) 4732 (*state)->timeout = PFTM_TCP_OPENING; 4733 else if (src->state >= TCPS_CLOSING || 4734 dst->state >= TCPS_CLOSING) 4735 (*state)->timeout = PFTM_TCP_CLOSING; 4736 else 4737 (*state)->timeout = PFTM_TCP_ESTABLISHED; 4738 4739 return (PF_PASS); 4740 } 4741 4742 static int 4743 pf_synproxy(struct pf_pdesc *pd, struct pf_kstate **state, u_short *reason) 4744 { 4745 struct pf_state_key *sk = (*state)->key[pd->didx]; 4746 struct tcphdr *th = &pd->hdr.tcp; 4747 4748 if ((*state)->src.state == PF_TCPS_PROXY_SRC) { 4749 if (pd->dir != (*state)->direction) { 4750 REASON_SET(reason, PFRES_SYNPROXY); 4751 return (PF_SYNPROXY_DROP); 4752 } 4753 if (th->th_flags & TH_SYN) { 4754 if (ntohl(th->th_seq) != (*state)->src.seqlo) { 4755 REASON_SET(reason, PFRES_SYNPROXY); 4756 return (PF_DROP); 4757 } 4758 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 4759 pd->src, th->th_dport, th->th_sport, 4760 (*state)->src.seqhi, ntohl(th->th_seq) + 1, 4761 TH_SYN|TH_ACK, 0, (*state)->src.mss, 0, 1, 0); 4762 REASON_SET(reason, PFRES_SYNPROXY); 4763 return (PF_SYNPROXY_DROP); 4764 } else if ((th->th_flags & (TH_ACK|TH_RST|TH_FIN)) != TH_ACK || 4765 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 4766 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 4767 REASON_SET(reason, PFRES_SYNPROXY); 4768 return (PF_DROP); 4769 } else if ((*state)->src_node != NULL && 4770 pf_src_connlimit(state)) { 4771 REASON_SET(reason, PFRES_SRCLIMIT); 4772 return (PF_DROP); 4773 } else 4774 pf_set_protostate(*state, PF_PEER_SRC, 4775 PF_TCPS_PROXY_DST); 4776 } 4777 if ((*state)->src.state == PF_TCPS_PROXY_DST) { 4778 if (pd->dir == (*state)->direction) { 4779 if (((th->th_flags & (TH_SYN|TH_ACK)) != TH_ACK) || 4780 (ntohl(th->th_ack) != (*state)->src.seqhi + 1) || 4781 (ntohl(th->th_seq) != (*state)->src.seqlo + 1)) { 4782 REASON_SET(reason, PFRES_SYNPROXY); 4783 return (PF_DROP); 4784 } 4785 (*state)->src.max_win = MAX(ntohs(th->th_win), 1); 4786 if ((*state)->dst.seqhi == 1) 4787 (*state)->dst.seqhi = htonl(arc4random()); 4788 pf_send_tcp((*state)->rule.ptr, pd->af, 4789 &sk->addr[pd->sidx], &sk->addr[pd->didx], 4790 sk->port[pd->sidx], sk->port[pd->didx], 4791 (*state)->dst.seqhi, 0, TH_SYN, 0, 4792 (*state)->src.mss, 0, 0, (*state)->tag); 4793 REASON_SET(reason, PFRES_SYNPROXY); 4794 return (PF_SYNPROXY_DROP); 4795 } else if (((th->th_flags & (TH_SYN|TH_ACK)) != 4796 (TH_SYN|TH_ACK)) || 4797 (ntohl(th->th_ack) != (*state)->dst.seqhi + 1)) { 4798 REASON_SET(reason, PFRES_SYNPROXY); 4799 return (PF_DROP); 4800 } else { 4801 (*state)->dst.max_win = MAX(ntohs(th->th_win), 1); 4802 (*state)->dst.seqlo = ntohl(th->th_seq); 4803 pf_send_tcp((*state)->rule.ptr, pd->af, pd->dst, 4804 pd->src, th->th_dport, th->th_sport, 4805 ntohl(th->th_ack), ntohl(th->th_seq) + 1, 4806 TH_ACK, (*state)->src.max_win, 0, 0, 0, 4807 (*state)->tag); 4808 pf_send_tcp((*state)->rule.ptr, pd->af, 4809 &sk->addr[pd->sidx], &sk->addr[pd->didx], 4810 sk->port[pd->sidx], sk->port[pd->didx], 4811 (*state)->src.seqhi + 1, (*state)->src.seqlo + 1, 4812 TH_ACK, (*state)->dst.max_win, 0, 0, 1, 0); 4813 (*state)->src.seqdiff = (*state)->dst.seqhi - 4814 (*state)->src.seqlo; 4815 (*state)->dst.seqdiff = (*state)->src.seqhi - 4816 (*state)->dst.seqlo; 4817 (*state)->src.seqhi = (*state)->src.seqlo + 4818 (*state)->dst.max_win; 4819 (*state)->dst.seqhi = (*state)->dst.seqlo + 4820 (*state)->src.max_win; 4821 (*state)->src.wscale = (*state)->dst.wscale = 0; 4822 pf_set_protostate(*state, PF_PEER_BOTH, 4823 TCPS_ESTABLISHED); 4824 REASON_SET(reason, PFRES_SYNPROXY); 4825 return (PF_SYNPROXY_DROP); 4826 } 4827 } 4828 4829 return (PF_PASS); 4830 } 4831 4832 static int 4833 pf_test_state_tcp(struct pf_kstate **state, int direction, struct pfi_kkif *kif, 4834 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, 4835 u_short *reason) 4836 { 4837 struct pf_state_key_cmp key; 4838 struct tcphdr *th = &pd->hdr.tcp; 4839 int copyback = 0; 4840 int action; 4841 struct pf_state_peer *src, *dst; 4842 struct pf_state_key *sk; 4843 4844 bzero(&key, sizeof(key)); 4845 key.af = pd->af; 4846 key.proto = IPPROTO_TCP; 4847 if (direction == PF_IN) { /* wire side, straight */ 4848 PF_ACPY(&key.addr[0], pd->src, key.af); 4849 PF_ACPY(&key.addr[1], pd->dst, key.af); 4850 key.port[0] = th->th_sport; 4851 key.port[1] = th->th_dport; 4852 } else { /* stack side, reverse */ 4853 PF_ACPY(&key.addr[1], pd->src, key.af); 4854 PF_ACPY(&key.addr[0], pd->dst, key.af); 4855 key.port[1] = th->th_sport; 4856 key.port[0] = th->th_dport; 4857 } 4858 4859 STATE_LOOKUP(kif, &key, direction, *state, pd); 4860 4861 if (direction == (*state)->direction) { 4862 src = &(*state)->src; 4863 dst = &(*state)->dst; 4864 } else { 4865 src = &(*state)->dst; 4866 dst = &(*state)->src; 4867 } 4868 4869 sk = (*state)->key[pd->didx]; 4870 4871 if ((action = pf_synproxy(pd, state, reason)) != PF_PASS) 4872 return (action); 4873 4874 if (((th->th_flags & (TH_SYN|TH_ACK)) == TH_SYN) && 4875 dst->state >= TCPS_FIN_WAIT_2 && 4876 src->state >= TCPS_FIN_WAIT_2) { 4877 if (V_pf_status.debug >= PF_DEBUG_MISC) { 4878 printf("pf: state reuse "); 4879 pf_print_state(*state); 4880 pf_print_flags(th->th_flags); 4881 printf("\n"); 4882 } 4883 /* XXX make sure it's the same direction ?? */ 4884 pf_set_protostate(*state, PF_PEER_BOTH, TCPS_CLOSED); 4885 pf_unlink_state(*state, PF_ENTER_LOCKED); 4886 *state = NULL; 4887 return (PF_DROP); 4888 } 4889 4890 if ((*state)->state_flags & PFSTATE_SLOPPY) { 4891 if (pf_tcp_track_sloppy(state, pd, reason) == PF_DROP) 4892 return (PF_DROP); 4893 } else { 4894 if (pf_tcp_track_full(state, kif, m, off, pd, reason, 4895 ©back) == PF_DROP) 4896 return (PF_DROP); 4897 } 4898 4899 /* translate source/destination address, if necessary */ 4900 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4901 struct pf_state_key *nk = (*state)->key[pd->didx]; 4902 4903 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 4904 nk->port[pd->sidx] != th->th_sport) 4905 pf_change_ap(m, pd->src, &th->th_sport, 4906 pd->ip_sum, &th->th_sum, &nk->addr[pd->sidx], 4907 nk->port[pd->sidx], 0, pd->af); 4908 4909 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 4910 nk->port[pd->didx] != th->th_dport) 4911 pf_change_ap(m, pd->dst, &th->th_dport, 4912 pd->ip_sum, &th->th_sum, &nk->addr[pd->didx], 4913 nk->port[pd->didx], 0, pd->af); 4914 copyback = 1; 4915 } 4916 4917 /* Copyback sequence modulation or stateful scrub changes if needed */ 4918 if (copyback) 4919 m_copyback(m, off, sizeof(*th), (caddr_t)th); 4920 4921 return (PF_PASS); 4922 } 4923 4924 static int 4925 pf_test_state_udp(struct pf_kstate **state, int direction, struct pfi_kkif *kif, 4926 struct mbuf *m, int off, void *h, struct pf_pdesc *pd) 4927 { 4928 struct pf_state_peer *src, *dst; 4929 struct pf_state_key_cmp key; 4930 struct udphdr *uh = &pd->hdr.udp; 4931 uint8_t psrc, pdst; 4932 4933 bzero(&key, sizeof(key)); 4934 key.af = pd->af; 4935 key.proto = IPPROTO_UDP; 4936 if (direction == PF_IN) { /* wire side, straight */ 4937 PF_ACPY(&key.addr[0], pd->src, key.af); 4938 PF_ACPY(&key.addr[1], pd->dst, key.af); 4939 key.port[0] = uh->uh_sport; 4940 key.port[1] = uh->uh_dport; 4941 } else { /* stack side, reverse */ 4942 PF_ACPY(&key.addr[1], pd->src, key.af); 4943 PF_ACPY(&key.addr[0], pd->dst, key.af); 4944 key.port[1] = uh->uh_sport; 4945 key.port[0] = uh->uh_dport; 4946 } 4947 4948 STATE_LOOKUP(kif, &key, direction, *state, pd); 4949 4950 if (direction == (*state)->direction) { 4951 src = &(*state)->src; 4952 dst = &(*state)->dst; 4953 psrc = PF_PEER_SRC; 4954 pdst = PF_PEER_DST; 4955 } else { 4956 src = &(*state)->dst; 4957 dst = &(*state)->src; 4958 psrc = PF_PEER_DST; 4959 pdst = PF_PEER_SRC; 4960 } 4961 4962 /* update states */ 4963 if (src->state < PFUDPS_SINGLE) 4964 pf_set_protostate(*state, psrc, PFUDPS_SINGLE); 4965 if (dst->state == PFUDPS_SINGLE) 4966 pf_set_protostate(*state, pdst, PFUDPS_MULTIPLE); 4967 4968 /* update expire time */ 4969 (*state)->expire = time_uptime; 4970 if (src->state == PFUDPS_MULTIPLE && dst->state == PFUDPS_MULTIPLE) 4971 (*state)->timeout = PFTM_UDP_MULTIPLE; 4972 else 4973 (*state)->timeout = PFTM_UDP_SINGLE; 4974 4975 /* translate source/destination address, if necessary */ 4976 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 4977 struct pf_state_key *nk = (*state)->key[pd->didx]; 4978 4979 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], pd->af) || 4980 nk->port[pd->sidx] != uh->uh_sport) 4981 pf_change_ap(m, pd->src, &uh->uh_sport, pd->ip_sum, 4982 &uh->uh_sum, &nk->addr[pd->sidx], 4983 nk->port[pd->sidx], 1, pd->af); 4984 4985 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], pd->af) || 4986 nk->port[pd->didx] != uh->uh_dport) 4987 pf_change_ap(m, pd->dst, &uh->uh_dport, pd->ip_sum, 4988 &uh->uh_sum, &nk->addr[pd->didx], 4989 nk->port[pd->didx], 1, pd->af); 4990 m_copyback(m, off, sizeof(*uh), (caddr_t)uh); 4991 } 4992 4993 return (PF_PASS); 4994 } 4995 4996 static int 4997 pf_test_state_icmp(struct pf_kstate **state, int direction, struct pfi_kkif *kif, 4998 struct mbuf *m, int off, void *h, struct pf_pdesc *pd, u_short *reason) 4999 { 5000 struct pf_addr *saddr = pd->src, *daddr = pd->dst; 5001 u_int16_t icmpid = 0, *icmpsum; 5002 u_int8_t icmptype, icmpcode; 5003 int state_icmp = 0; 5004 struct pf_state_key_cmp key; 5005 5006 bzero(&key, sizeof(key)); 5007 switch (pd->proto) { 5008 #ifdef INET 5009 case IPPROTO_ICMP: 5010 icmptype = pd->hdr.icmp.icmp_type; 5011 icmpcode = pd->hdr.icmp.icmp_code; 5012 icmpid = pd->hdr.icmp.icmp_id; 5013 icmpsum = &pd->hdr.icmp.icmp_cksum; 5014 5015 if (icmptype == ICMP_UNREACH || 5016 icmptype == ICMP_SOURCEQUENCH || 5017 icmptype == ICMP_REDIRECT || 5018 icmptype == ICMP_TIMXCEED || 5019 icmptype == ICMP_PARAMPROB) 5020 state_icmp++; 5021 break; 5022 #endif /* INET */ 5023 #ifdef INET6 5024 case IPPROTO_ICMPV6: 5025 icmptype = pd->hdr.icmp6.icmp6_type; 5026 icmpcode = pd->hdr.icmp6.icmp6_code; 5027 icmpid = pd->hdr.icmp6.icmp6_id; 5028 icmpsum = &pd->hdr.icmp6.icmp6_cksum; 5029 5030 if (icmptype == ICMP6_DST_UNREACH || 5031 icmptype == ICMP6_PACKET_TOO_BIG || 5032 icmptype == ICMP6_TIME_EXCEEDED || 5033 icmptype == ICMP6_PARAM_PROB) 5034 state_icmp++; 5035 break; 5036 #endif /* INET6 */ 5037 } 5038 5039 if (!state_icmp) { 5040 /* 5041 * ICMP query/reply message not related to a TCP/UDP packet. 5042 * Search for an ICMP state. 5043 */ 5044 key.af = pd->af; 5045 key.proto = pd->proto; 5046 key.port[0] = key.port[1] = icmpid; 5047 if (direction == PF_IN) { /* wire side, straight */ 5048 PF_ACPY(&key.addr[0], pd->src, key.af); 5049 PF_ACPY(&key.addr[1], pd->dst, key.af); 5050 } else { /* stack side, reverse */ 5051 PF_ACPY(&key.addr[1], pd->src, key.af); 5052 PF_ACPY(&key.addr[0], pd->dst, key.af); 5053 } 5054 5055 STATE_LOOKUP(kif, &key, direction, *state, pd); 5056 5057 (*state)->expire = time_uptime; 5058 (*state)->timeout = PFTM_ICMP_ERROR_REPLY; 5059 5060 /* translate source/destination address, if necessary */ 5061 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5062 struct pf_state_key *nk = (*state)->key[pd->didx]; 5063 5064 switch (pd->af) { 5065 #ifdef INET 5066 case AF_INET: 5067 if (PF_ANEQ(pd->src, 5068 &nk->addr[pd->sidx], AF_INET)) 5069 pf_change_a(&saddr->v4.s_addr, 5070 pd->ip_sum, 5071 nk->addr[pd->sidx].v4.s_addr, 0); 5072 5073 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], 5074 AF_INET)) 5075 pf_change_a(&daddr->v4.s_addr, 5076 pd->ip_sum, 5077 nk->addr[pd->didx].v4.s_addr, 0); 5078 5079 if (nk->port[0] != 5080 pd->hdr.icmp.icmp_id) { 5081 pd->hdr.icmp.icmp_cksum = 5082 pf_cksum_fixup( 5083 pd->hdr.icmp.icmp_cksum, icmpid, 5084 nk->port[pd->sidx], 0); 5085 pd->hdr.icmp.icmp_id = 5086 nk->port[pd->sidx]; 5087 } 5088 5089 m_copyback(m, off, ICMP_MINLEN, 5090 (caddr_t )&pd->hdr.icmp); 5091 break; 5092 #endif /* INET */ 5093 #ifdef INET6 5094 case AF_INET6: 5095 if (PF_ANEQ(pd->src, 5096 &nk->addr[pd->sidx], AF_INET6)) 5097 pf_change_a6(saddr, 5098 &pd->hdr.icmp6.icmp6_cksum, 5099 &nk->addr[pd->sidx], 0); 5100 5101 if (PF_ANEQ(pd->dst, 5102 &nk->addr[pd->didx], AF_INET6)) 5103 pf_change_a6(daddr, 5104 &pd->hdr.icmp6.icmp6_cksum, 5105 &nk->addr[pd->didx], 0); 5106 5107 m_copyback(m, off, sizeof(struct icmp6_hdr), 5108 (caddr_t )&pd->hdr.icmp6); 5109 break; 5110 #endif /* INET6 */ 5111 } 5112 } 5113 return (PF_PASS); 5114 5115 } else { 5116 /* 5117 * ICMP error message in response to a TCP/UDP packet. 5118 * Extract the inner TCP/UDP header and search for that state. 5119 */ 5120 5121 struct pf_pdesc pd2; 5122 bzero(&pd2, sizeof pd2); 5123 #ifdef INET 5124 struct ip h2; 5125 #endif /* INET */ 5126 #ifdef INET6 5127 struct ip6_hdr h2_6; 5128 int terminal = 0; 5129 #endif /* INET6 */ 5130 int ipoff2 = 0; 5131 int off2 = 0; 5132 5133 pd2.af = pd->af; 5134 /* Payload packet is from the opposite direction. */ 5135 pd2.sidx = (direction == PF_IN) ? 1 : 0; 5136 pd2.didx = (direction == PF_IN) ? 0 : 1; 5137 switch (pd->af) { 5138 #ifdef INET 5139 case AF_INET: 5140 /* offset of h2 in mbuf chain */ 5141 ipoff2 = off + ICMP_MINLEN; 5142 5143 if (!pf_pull_hdr(m, ipoff2, &h2, sizeof(h2), 5144 NULL, reason, pd2.af)) { 5145 DPFPRINTF(PF_DEBUG_MISC, 5146 ("pf: ICMP error message too short " 5147 "(ip)\n")); 5148 return (PF_DROP); 5149 } 5150 /* 5151 * ICMP error messages don't refer to non-first 5152 * fragments 5153 */ 5154 if (h2.ip_off & htons(IP_OFFMASK)) { 5155 REASON_SET(reason, PFRES_FRAG); 5156 return (PF_DROP); 5157 } 5158 5159 /* offset of protocol header that follows h2 */ 5160 off2 = ipoff2 + (h2.ip_hl << 2); 5161 5162 pd2.proto = h2.ip_p; 5163 pd2.src = (struct pf_addr *)&h2.ip_src; 5164 pd2.dst = (struct pf_addr *)&h2.ip_dst; 5165 pd2.ip_sum = &h2.ip_sum; 5166 break; 5167 #endif /* INET */ 5168 #ifdef INET6 5169 case AF_INET6: 5170 ipoff2 = off + sizeof(struct icmp6_hdr); 5171 5172 if (!pf_pull_hdr(m, ipoff2, &h2_6, sizeof(h2_6), 5173 NULL, reason, pd2.af)) { 5174 DPFPRINTF(PF_DEBUG_MISC, 5175 ("pf: ICMP error message too short " 5176 "(ip6)\n")); 5177 return (PF_DROP); 5178 } 5179 pd2.proto = h2_6.ip6_nxt; 5180 pd2.src = (struct pf_addr *)&h2_6.ip6_src; 5181 pd2.dst = (struct pf_addr *)&h2_6.ip6_dst; 5182 pd2.ip_sum = NULL; 5183 off2 = ipoff2 + sizeof(h2_6); 5184 do { 5185 switch (pd2.proto) { 5186 case IPPROTO_FRAGMENT: 5187 /* 5188 * ICMPv6 error messages for 5189 * non-first fragments 5190 */ 5191 REASON_SET(reason, PFRES_FRAG); 5192 return (PF_DROP); 5193 case IPPROTO_AH: 5194 case IPPROTO_HOPOPTS: 5195 case IPPROTO_ROUTING: 5196 case IPPROTO_DSTOPTS: { 5197 /* get next header and header length */ 5198 struct ip6_ext opt6; 5199 5200 if (!pf_pull_hdr(m, off2, &opt6, 5201 sizeof(opt6), NULL, reason, 5202 pd2.af)) { 5203 DPFPRINTF(PF_DEBUG_MISC, 5204 ("pf: ICMPv6 short opt\n")); 5205 return (PF_DROP); 5206 } 5207 if (pd2.proto == IPPROTO_AH) 5208 off2 += (opt6.ip6e_len + 2) * 4; 5209 else 5210 off2 += (opt6.ip6e_len + 1) * 8; 5211 pd2.proto = opt6.ip6e_nxt; 5212 /* goto the next header */ 5213 break; 5214 } 5215 default: 5216 terminal++; 5217 break; 5218 } 5219 } while (!terminal); 5220 break; 5221 #endif /* INET6 */ 5222 } 5223 5224 if (PF_ANEQ(pd->dst, pd2.src, pd->af)) { 5225 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5226 printf("pf: BAD ICMP %d:%d outer dst: ", 5227 icmptype, icmpcode); 5228 pf_print_host(pd->src, 0, pd->af); 5229 printf(" -> "); 5230 pf_print_host(pd->dst, 0, pd->af); 5231 printf(" inner src: "); 5232 pf_print_host(pd2.src, 0, pd2.af); 5233 printf(" -> "); 5234 pf_print_host(pd2.dst, 0, pd2.af); 5235 printf("\n"); 5236 } 5237 REASON_SET(reason, PFRES_BADSTATE); 5238 return (PF_DROP); 5239 } 5240 5241 switch (pd2.proto) { 5242 case IPPROTO_TCP: { 5243 struct tcphdr th; 5244 u_int32_t seq; 5245 struct pf_state_peer *src, *dst; 5246 u_int8_t dws; 5247 int copyback = 0; 5248 5249 /* 5250 * Only the first 8 bytes of the TCP header can be 5251 * expected. Don't access any TCP header fields after 5252 * th_seq, an ackskew test is not possible. 5253 */ 5254 if (!pf_pull_hdr(m, off2, &th, 8, NULL, reason, 5255 pd2.af)) { 5256 DPFPRINTF(PF_DEBUG_MISC, 5257 ("pf: ICMP error message too short " 5258 "(tcp)\n")); 5259 return (PF_DROP); 5260 } 5261 5262 key.af = pd2.af; 5263 key.proto = IPPROTO_TCP; 5264 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5265 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5266 key.port[pd2.sidx] = th.th_sport; 5267 key.port[pd2.didx] = th.th_dport; 5268 5269 STATE_LOOKUP(kif, &key, direction, *state, pd); 5270 5271 if (direction == (*state)->direction) { 5272 src = &(*state)->dst; 5273 dst = &(*state)->src; 5274 } else { 5275 src = &(*state)->src; 5276 dst = &(*state)->dst; 5277 } 5278 5279 if (src->wscale && dst->wscale) 5280 dws = dst->wscale & PF_WSCALE_MASK; 5281 else 5282 dws = 0; 5283 5284 /* Demodulate sequence number */ 5285 seq = ntohl(th.th_seq) - src->seqdiff; 5286 if (src->seqdiff) { 5287 pf_change_a(&th.th_seq, icmpsum, 5288 htonl(seq), 0); 5289 copyback = 1; 5290 } 5291 5292 if (!((*state)->state_flags & PFSTATE_SLOPPY) && 5293 (!SEQ_GEQ(src->seqhi, seq) || 5294 !SEQ_GEQ(seq, src->seqlo - (dst->max_win << dws)))) { 5295 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5296 printf("pf: BAD ICMP %d:%d ", 5297 icmptype, icmpcode); 5298 pf_print_host(pd->src, 0, pd->af); 5299 printf(" -> "); 5300 pf_print_host(pd->dst, 0, pd->af); 5301 printf(" state: "); 5302 pf_print_state(*state); 5303 printf(" seq=%u\n", seq); 5304 } 5305 REASON_SET(reason, PFRES_BADSTATE); 5306 return (PF_DROP); 5307 } else { 5308 if (V_pf_status.debug >= PF_DEBUG_MISC) { 5309 printf("pf: OK ICMP %d:%d ", 5310 icmptype, icmpcode); 5311 pf_print_host(pd->src, 0, pd->af); 5312 printf(" -> "); 5313 pf_print_host(pd->dst, 0, pd->af); 5314 printf(" state: "); 5315 pf_print_state(*state); 5316 printf(" seq=%u\n", seq); 5317 } 5318 } 5319 5320 /* translate source/destination address, if necessary */ 5321 if ((*state)->key[PF_SK_WIRE] != 5322 (*state)->key[PF_SK_STACK]) { 5323 struct pf_state_key *nk = 5324 (*state)->key[pd->didx]; 5325 5326 if (PF_ANEQ(pd2.src, 5327 &nk->addr[pd2.sidx], pd2.af) || 5328 nk->port[pd2.sidx] != th.th_sport) 5329 pf_change_icmp(pd2.src, &th.th_sport, 5330 daddr, &nk->addr[pd2.sidx], 5331 nk->port[pd2.sidx], NULL, 5332 pd2.ip_sum, icmpsum, 5333 pd->ip_sum, 0, pd2.af); 5334 5335 if (PF_ANEQ(pd2.dst, 5336 &nk->addr[pd2.didx], pd2.af) || 5337 nk->port[pd2.didx] != th.th_dport) 5338 pf_change_icmp(pd2.dst, &th.th_dport, 5339 saddr, &nk->addr[pd2.didx], 5340 nk->port[pd2.didx], NULL, 5341 pd2.ip_sum, icmpsum, 5342 pd->ip_sum, 0, pd2.af); 5343 copyback = 1; 5344 } 5345 5346 if (copyback) { 5347 switch (pd2.af) { 5348 #ifdef INET 5349 case AF_INET: 5350 m_copyback(m, off, ICMP_MINLEN, 5351 (caddr_t )&pd->hdr.icmp); 5352 m_copyback(m, ipoff2, sizeof(h2), 5353 (caddr_t )&h2); 5354 break; 5355 #endif /* INET */ 5356 #ifdef INET6 5357 case AF_INET6: 5358 m_copyback(m, off, 5359 sizeof(struct icmp6_hdr), 5360 (caddr_t )&pd->hdr.icmp6); 5361 m_copyback(m, ipoff2, sizeof(h2_6), 5362 (caddr_t )&h2_6); 5363 break; 5364 #endif /* INET6 */ 5365 } 5366 m_copyback(m, off2, 8, (caddr_t)&th); 5367 } 5368 5369 return (PF_PASS); 5370 break; 5371 } 5372 case IPPROTO_UDP: { 5373 struct udphdr uh; 5374 5375 if (!pf_pull_hdr(m, off2, &uh, sizeof(uh), 5376 NULL, reason, pd2.af)) { 5377 DPFPRINTF(PF_DEBUG_MISC, 5378 ("pf: ICMP error message too short " 5379 "(udp)\n")); 5380 return (PF_DROP); 5381 } 5382 5383 key.af = pd2.af; 5384 key.proto = IPPROTO_UDP; 5385 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5386 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5387 key.port[pd2.sidx] = uh.uh_sport; 5388 key.port[pd2.didx] = uh.uh_dport; 5389 5390 STATE_LOOKUP(kif, &key, direction, *state, pd); 5391 5392 /* translate source/destination address, if necessary */ 5393 if ((*state)->key[PF_SK_WIRE] != 5394 (*state)->key[PF_SK_STACK]) { 5395 struct pf_state_key *nk = 5396 (*state)->key[pd->didx]; 5397 5398 if (PF_ANEQ(pd2.src, 5399 &nk->addr[pd2.sidx], pd2.af) || 5400 nk->port[pd2.sidx] != uh.uh_sport) 5401 pf_change_icmp(pd2.src, &uh.uh_sport, 5402 daddr, &nk->addr[pd2.sidx], 5403 nk->port[pd2.sidx], &uh.uh_sum, 5404 pd2.ip_sum, icmpsum, 5405 pd->ip_sum, 1, pd2.af); 5406 5407 if (PF_ANEQ(pd2.dst, 5408 &nk->addr[pd2.didx], pd2.af) || 5409 nk->port[pd2.didx] != uh.uh_dport) 5410 pf_change_icmp(pd2.dst, &uh.uh_dport, 5411 saddr, &nk->addr[pd2.didx], 5412 nk->port[pd2.didx], &uh.uh_sum, 5413 pd2.ip_sum, icmpsum, 5414 pd->ip_sum, 1, pd2.af); 5415 5416 switch (pd2.af) { 5417 #ifdef INET 5418 case AF_INET: 5419 m_copyback(m, off, ICMP_MINLEN, 5420 (caddr_t )&pd->hdr.icmp); 5421 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 5422 break; 5423 #endif /* INET */ 5424 #ifdef INET6 5425 case AF_INET6: 5426 m_copyback(m, off, 5427 sizeof(struct icmp6_hdr), 5428 (caddr_t )&pd->hdr.icmp6); 5429 m_copyback(m, ipoff2, sizeof(h2_6), 5430 (caddr_t )&h2_6); 5431 break; 5432 #endif /* INET6 */ 5433 } 5434 m_copyback(m, off2, sizeof(uh), (caddr_t)&uh); 5435 } 5436 return (PF_PASS); 5437 break; 5438 } 5439 #ifdef INET 5440 case IPPROTO_ICMP: { 5441 struct icmp iih; 5442 5443 if (!pf_pull_hdr(m, off2, &iih, ICMP_MINLEN, 5444 NULL, reason, pd2.af)) { 5445 DPFPRINTF(PF_DEBUG_MISC, 5446 ("pf: ICMP error message too short i" 5447 "(icmp)\n")); 5448 return (PF_DROP); 5449 } 5450 5451 key.af = pd2.af; 5452 key.proto = IPPROTO_ICMP; 5453 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5454 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5455 key.port[0] = key.port[1] = iih.icmp_id; 5456 5457 STATE_LOOKUP(kif, &key, direction, *state, pd); 5458 5459 /* translate source/destination address, if necessary */ 5460 if ((*state)->key[PF_SK_WIRE] != 5461 (*state)->key[PF_SK_STACK]) { 5462 struct pf_state_key *nk = 5463 (*state)->key[pd->didx]; 5464 5465 if (PF_ANEQ(pd2.src, 5466 &nk->addr[pd2.sidx], pd2.af) || 5467 nk->port[pd2.sidx] != iih.icmp_id) 5468 pf_change_icmp(pd2.src, &iih.icmp_id, 5469 daddr, &nk->addr[pd2.sidx], 5470 nk->port[pd2.sidx], NULL, 5471 pd2.ip_sum, icmpsum, 5472 pd->ip_sum, 0, AF_INET); 5473 5474 if (PF_ANEQ(pd2.dst, 5475 &nk->addr[pd2.didx], pd2.af) || 5476 nk->port[pd2.didx] != iih.icmp_id) 5477 pf_change_icmp(pd2.dst, &iih.icmp_id, 5478 saddr, &nk->addr[pd2.didx], 5479 nk->port[pd2.didx], NULL, 5480 pd2.ip_sum, icmpsum, 5481 pd->ip_sum, 0, AF_INET); 5482 5483 m_copyback(m, off, ICMP_MINLEN, (caddr_t)&pd->hdr.icmp); 5484 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 5485 m_copyback(m, off2, ICMP_MINLEN, (caddr_t)&iih); 5486 } 5487 return (PF_PASS); 5488 break; 5489 } 5490 #endif /* INET */ 5491 #ifdef INET6 5492 case IPPROTO_ICMPV6: { 5493 struct icmp6_hdr iih; 5494 5495 if (!pf_pull_hdr(m, off2, &iih, 5496 sizeof(struct icmp6_hdr), NULL, reason, pd2.af)) { 5497 DPFPRINTF(PF_DEBUG_MISC, 5498 ("pf: ICMP error message too short " 5499 "(icmp6)\n")); 5500 return (PF_DROP); 5501 } 5502 5503 key.af = pd2.af; 5504 key.proto = IPPROTO_ICMPV6; 5505 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5506 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5507 key.port[0] = key.port[1] = iih.icmp6_id; 5508 5509 STATE_LOOKUP(kif, &key, direction, *state, pd); 5510 5511 /* translate source/destination address, if necessary */ 5512 if ((*state)->key[PF_SK_WIRE] != 5513 (*state)->key[PF_SK_STACK]) { 5514 struct pf_state_key *nk = 5515 (*state)->key[pd->didx]; 5516 5517 if (PF_ANEQ(pd2.src, 5518 &nk->addr[pd2.sidx], pd2.af) || 5519 nk->port[pd2.sidx] != iih.icmp6_id) 5520 pf_change_icmp(pd2.src, &iih.icmp6_id, 5521 daddr, &nk->addr[pd2.sidx], 5522 nk->port[pd2.sidx], NULL, 5523 pd2.ip_sum, icmpsum, 5524 pd->ip_sum, 0, AF_INET6); 5525 5526 if (PF_ANEQ(pd2.dst, 5527 &nk->addr[pd2.didx], pd2.af) || 5528 nk->port[pd2.didx] != iih.icmp6_id) 5529 pf_change_icmp(pd2.dst, &iih.icmp6_id, 5530 saddr, &nk->addr[pd2.didx], 5531 nk->port[pd2.didx], NULL, 5532 pd2.ip_sum, icmpsum, 5533 pd->ip_sum, 0, AF_INET6); 5534 5535 m_copyback(m, off, sizeof(struct icmp6_hdr), 5536 (caddr_t)&pd->hdr.icmp6); 5537 m_copyback(m, ipoff2, sizeof(h2_6), (caddr_t)&h2_6); 5538 m_copyback(m, off2, sizeof(struct icmp6_hdr), 5539 (caddr_t)&iih); 5540 } 5541 return (PF_PASS); 5542 break; 5543 } 5544 #endif /* INET6 */ 5545 default: { 5546 key.af = pd2.af; 5547 key.proto = pd2.proto; 5548 PF_ACPY(&key.addr[pd2.sidx], pd2.src, key.af); 5549 PF_ACPY(&key.addr[pd2.didx], pd2.dst, key.af); 5550 key.port[0] = key.port[1] = 0; 5551 5552 STATE_LOOKUP(kif, &key, direction, *state, pd); 5553 5554 /* translate source/destination address, if necessary */ 5555 if ((*state)->key[PF_SK_WIRE] != 5556 (*state)->key[PF_SK_STACK]) { 5557 struct pf_state_key *nk = 5558 (*state)->key[pd->didx]; 5559 5560 if (PF_ANEQ(pd2.src, 5561 &nk->addr[pd2.sidx], pd2.af)) 5562 pf_change_icmp(pd2.src, NULL, daddr, 5563 &nk->addr[pd2.sidx], 0, NULL, 5564 pd2.ip_sum, icmpsum, 5565 pd->ip_sum, 0, pd2.af); 5566 5567 if (PF_ANEQ(pd2.dst, 5568 &nk->addr[pd2.didx], pd2.af)) 5569 pf_change_icmp(pd2.dst, NULL, saddr, 5570 &nk->addr[pd2.didx], 0, NULL, 5571 pd2.ip_sum, icmpsum, 5572 pd->ip_sum, 0, pd2.af); 5573 5574 switch (pd2.af) { 5575 #ifdef INET 5576 case AF_INET: 5577 m_copyback(m, off, ICMP_MINLEN, 5578 (caddr_t)&pd->hdr.icmp); 5579 m_copyback(m, ipoff2, sizeof(h2), (caddr_t)&h2); 5580 break; 5581 #endif /* INET */ 5582 #ifdef INET6 5583 case AF_INET6: 5584 m_copyback(m, off, 5585 sizeof(struct icmp6_hdr), 5586 (caddr_t )&pd->hdr.icmp6); 5587 m_copyback(m, ipoff2, sizeof(h2_6), 5588 (caddr_t )&h2_6); 5589 break; 5590 #endif /* INET6 */ 5591 } 5592 } 5593 return (PF_PASS); 5594 break; 5595 } 5596 } 5597 } 5598 } 5599 5600 static int 5601 pf_test_state_other(struct pf_kstate **state, int direction, struct pfi_kkif *kif, 5602 struct mbuf *m, struct pf_pdesc *pd) 5603 { 5604 struct pf_state_peer *src, *dst; 5605 struct pf_state_key_cmp key; 5606 uint8_t psrc, pdst; 5607 5608 bzero(&key, sizeof(key)); 5609 key.af = pd->af; 5610 key.proto = pd->proto; 5611 if (direction == PF_IN) { 5612 PF_ACPY(&key.addr[0], pd->src, key.af); 5613 PF_ACPY(&key.addr[1], pd->dst, key.af); 5614 key.port[0] = key.port[1] = 0; 5615 } else { 5616 PF_ACPY(&key.addr[1], pd->src, key.af); 5617 PF_ACPY(&key.addr[0], pd->dst, key.af); 5618 key.port[1] = key.port[0] = 0; 5619 } 5620 5621 STATE_LOOKUP(kif, &key, direction, *state, pd); 5622 5623 if (direction == (*state)->direction) { 5624 src = &(*state)->src; 5625 dst = &(*state)->dst; 5626 psrc = PF_PEER_SRC; 5627 pdst = PF_PEER_DST; 5628 } else { 5629 src = &(*state)->dst; 5630 dst = &(*state)->src; 5631 psrc = PF_PEER_DST; 5632 pdst = PF_PEER_SRC; 5633 } 5634 5635 /* update states */ 5636 if (src->state < PFOTHERS_SINGLE) 5637 pf_set_protostate(*state, psrc, PFOTHERS_SINGLE); 5638 if (dst->state == PFOTHERS_SINGLE) 5639 pf_set_protostate(*state, pdst, PFOTHERS_MULTIPLE); 5640 5641 /* update expire time */ 5642 (*state)->expire = time_uptime; 5643 if (src->state == PFOTHERS_MULTIPLE && dst->state == PFOTHERS_MULTIPLE) 5644 (*state)->timeout = PFTM_OTHER_MULTIPLE; 5645 else 5646 (*state)->timeout = PFTM_OTHER_SINGLE; 5647 5648 /* translate source/destination address, if necessary */ 5649 if ((*state)->key[PF_SK_WIRE] != (*state)->key[PF_SK_STACK]) { 5650 struct pf_state_key *nk = (*state)->key[pd->didx]; 5651 5652 KASSERT(nk, ("%s: nk is null", __func__)); 5653 KASSERT(pd, ("%s: pd is null", __func__)); 5654 KASSERT(pd->src, ("%s: pd->src is null", __func__)); 5655 KASSERT(pd->dst, ("%s: pd->dst is null", __func__)); 5656 switch (pd->af) { 5657 #ifdef INET 5658 case AF_INET: 5659 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 5660 pf_change_a(&pd->src->v4.s_addr, 5661 pd->ip_sum, 5662 nk->addr[pd->sidx].v4.s_addr, 5663 0); 5664 5665 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 5666 pf_change_a(&pd->dst->v4.s_addr, 5667 pd->ip_sum, 5668 nk->addr[pd->didx].v4.s_addr, 5669 0); 5670 5671 break; 5672 #endif /* INET */ 5673 #ifdef INET6 5674 case AF_INET6: 5675 if (PF_ANEQ(pd->src, &nk->addr[pd->sidx], AF_INET)) 5676 PF_ACPY(pd->src, &nk->addr[pd->sidx], pd->af); 5677 5678 if (PF_ANEQ(pd->dst, &nk->addr[pd->didx], AF_INET)) 5679 PF_ACPY(pd->dst, &nk->addr[pd->didx], pd->af); 5680 #endif /* INET6 */ 5681 } 5682 } 5683 return (PF_PASS); 5684 } 5685 5686 /* 5687 * ipoff and off are measured from the start of the mbuf chain. 5688 * h must be at "ipoff" on the mbuf chain. 5689 */ 5690 void * 5691 pf_pull_hdr(struct mbuf *m, int off, void *p, int len, 5692 u_short *actionp, u_short *reasonp, sa_family_t af) 5693 { 5694 switch (af) { 5695 #ifdef INET 5696 case AF_INET: { 5697 struct ip *h = mtod(m, struct ip *); 5698 u_int16_t fragoff = (ntohs(h->ip_off) & IP_OFFMASK) << 3; 5699 5700 if (fragoff) { 5701 if (fragoff >= len) 5702 ACTION_SET(actionp, PF_PASS); 5703 else { 5704 ACTION_SET(actionp, PF_DROP); 5705 REASON_SET(reasonp, PFRES_FRAG); 5706 } 5707 return (NULL); 5708 } 5709 if (m->m_pkthdr.len < off + len || 5710 ntohs(h->ip_len) < off + len) { 5711 ACTION_SET(actionp, PF_DROP); 5712 REASON_SET(reasonp, PFRES_SHORT); 5713 return (NULL); 5714 } 5715 break; 5716 } 5717 #endif /* INET */ 5718 #ifdef INET6 5719 case AF_INET6: { 5720 struct ip6_hdr *h = mtod(m, struct ip6_hdr *); 5721 5722 if (m->m_pkthdr.len < off + len || 5723 (ntohs(h->ip6_plen) + sizeof(struct ip6_hdr)) < 5724 (unsigned)(off + len)) { 5725 ACTION_SET(actionp, PF_DROP); 5726 REASON_SET(reasonp, PFRES_SHORT); 5727 return (NULL); 5728 } 5729 break; 5730 } 5731 #endif /* INET6 */ 5732 } 5733 m_copydata(m, off, len, p); 5734 return (p); 5735 } 5736 5737 int 5738 pf_routable(struct pf_addr *addr, sa_family_t af, struct pfi_kkif *kif, 5739 int rtableid) 5740 { 5741 struct ifnet *ifp; 5742 5743 /* 5744 * Skip check for addresses with embedded interface scope, 5745 * as they would always match anyway. 5746 */ 5747 if (af == AF_INET6 && IN6_IS_SCOPE_EMBED(&addr->v6)) 5748 return (1); 5749 5750 if (af != AF_INET && af != AF_INET6) 5751 return (0); 5752 5753 /* Skip checks for ipsec interfaces */ 5754 if (kif != NULL && kif->pfik_ifp->if_type == IFT_ENC) 5755 return (1); 5756 5757 ifp = (kif != NULL) ? kif->pfik_ifp : NULL; 5758 5759 switch (af) { 5760 #ifdef INET6 5761 case AF_INET6: 5762 return (fib6_check_urpf(rtableid, &addr->v6, 0, NHR_NONE, 5763 ifp)); 5764 #endif 5765 #ifdef INET 5766 case AF_INET: 5767 return (fib4_check_urpf(rtableid, addr->v4, 0, NHR_NONE, 5768 ifp)); 5769 #endif 5770 } 5771 5772 return (0); 5773 } 5774 5775 #ifdef INET 5776 static void 5777 pf_route(struct mbuf **m, struct pf_krule *r, int dir, struct ifnet *oifp, 5778 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 5779 { 5780 struct mbuf *m0, *m1; 5781 struct sockaddr_in dst; 5782 struct ip *ip; 5783 struct ifnet *ifp = NULL; 5784 struct pf_addr naddr; 5785 struct pf_ksrc_node *sn = NULL; 5786 int error = 0; 5787 uint16_t ip_len, ip_off; 5788 5789 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 5790 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", 5791 __func__)); 5792 5793 if ((pd->pf_mtag == NULL && 5794 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 5795 pd->pf_mtag->routed++ > 3) { 5796 m0 = *m; 5797 *m = NULL; 5798 goto bad_locked; 5799 } 5800 5801 if (r->rt == PF_DUPTO) { 5802 if ((pd->pf_mtag->flags & PF_DUPLICATED)) { 5803 if (s == NULL) { 5804 ifp = r->rpool.cur->kif ? 5805 r->rpool.cur->kif->pfik_ifp : NULL; 5806 } else { 5807 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 5808 PF_STATE_UNLOCK(s); 5809 } 5810 if (ifp == oifp) { 5811 /* When the 2nd interface is not skipped */ 5812 return; 5813 } else { 5814 m0 = *m; 5815 *m = NULL; 5816 goto bad; 5817 } 5818 } else { 5819 pd->pf_mtag->flags |= PF_DUPLICATED; 5820 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 5821 if (s) 5822 PF_STATE_UNLOCK(s); 5823 return; 5824 } 5825 } 5826 } else { 5827 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 5828 if (s) 5829 PF_STATE_UNLOCK(s); 5830 return; 5831 } 5832 m0 = *m; 5833 } 5834 5835 ip = mtod(m0, struct ip *); 5836 5837 bzero(&dst, sizeof(dst)); 5838 dst.sin_family = AF_INET; 5839 dst.sin_len = sizeof(dst); 5840 dst.sin_addr = ip->ip_dst; 5841 5842 bzero(&naddr, sizeof(naddr)); 5843 5844 if (TAILQ_EMPTY(&r->rpool.list)) { 5845 DPFPRINTF(PF_DEBUG_URGENT, 5846 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 5847 goto bad_locked; 5848 } 5849 if (s == NULL) { 5850 pf_map_addr(AF_INET, r, (struct pf_addr *)&ip->ip_src, 5851 &naddr, NULL, &sn); 5852 if (!PF_AZERO(&naddr, AF_INET)) 5853 dst.sin_addr.s_addr = naddr.v4.s_addr; 5854 ifp = r->rpool.cur->kif ? 5855 r->rpool.cur->kif->pfik_ifp : NULL; 5856 } else { 5857 if (!PF_AZERO(&s->rt_addr, AF_INET)) 5858 dst.sin_addr.s_addr = 5859 s->rt_addr.v4.s_addr; 5860 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 5861 PF_STATE_UNLOCK(s); 5862 } 5863 if (ifp == NULL) 5864 goto bad; 5865 5866 if (dir == PF_IN) { 5867 if (pf_test(PF_OUT, 0, ifp, &m0, inp) != PF_PASS) 5868 goto bad; 5869 else if (m0 == NULL) 5870 goto done; 5871 if (m0->m_len < sizeof(struct ip)) { 5872 DPFPRINTF(PF_DEBUG_URGENT, 5873 ("%s: m0->m_len < sizeof(struct ip)\n", __func__)); 5874 goto bad; 5875 } 5876 ip = mtod(m0, struct ip *); 5877 } 5878 5879 if (ifp->if_flags & IFF_LOOPBACK) 5880 m0->m_flags |= M_SKIP_FIREWALL; 5881 5882 ip_len = ntohs(ip->ip_len); 5883 ip_off = ntohs(ip->ip_off); 5884 5885 /* Copied from FreeBSD 10.0-CURRENT ip_output. */ 5886 m0->m_pkthdr.csum_flags |= CSUM_IP; 5887 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) { 5888 m0 = mb_unmapped_to_ext(m0); 5889 if (m0 == NULL) 5890 goto done; 5891 in_delayed_cksum(m0); 5892 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 5893 } 5894 #if defined(SCTP) || defined(SCTP_SUPPORT) 5895 if (m0->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) { 5896 m0 = mb_unmapped_to_ext(m0); 5897 if (m0 == NULL) 5898 goto done; 5899 sctp_delayed_cksum(m0, (uint32_t)(ip->ip_hl << 2)); 5900 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP; 5901 } 5902 #endif 5903 5904 /* 5905 * If small enough for interface, or the interface will take 5906 * care of the fragmentation for us, we can just send directly. 5907 */ 5908 if (ip_len <= ifp->if_mtu || 5909 (m0->m_pkthdr.csum_flags & ifp->if_hwassist & CSUM_TSO) != 0) { 5910 ip->ip_sum = 0; 5911 if (m0->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) { 5912 ip->ip_sum = in_cksum(m0, ip->ip_hl << 2); 5913 m0->m_pkthdr.csum_flags &= ~CSUM_IP; 5914 } 5915 m_clrprotoflags(m0); /* Avoid confusing lower layers. */ 5916 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL); 5917 goto done; 5918 } 5919 5920 /* Balk when DF bit is set or the interface didn't support TSO. */ 5921 if ((ip_off & IP_DF) || (m0->m_pkthdr.csum_flags & CSUM_TSO)) { 5922 error = EMSGSIZE; 5923 KMOD_IPSTAT_INC(ips_cantfrag); 5924 if (r->rt != PF_DUPTO) { 5925 icmp_error(m0, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 0, 5926 ifp->if_mtu); 5927 goto done; 5928 } else 5929 goto bad; 5930 } 5931 5932 error = ip_fragment(ip, &m0, ifp->if_mtu, ifp->if_hwassist); 5933 if (error) 5934 goto bad; 5935 5936 for (; m0; m0 = m1) { 5937 m1 = m0->m_nextpkt; 5938 m0->m_nextpkt = NULL; 5939 if (error == 0) { 5940 m_clrprotoflags(m0); 5941 error = (*ifp->if_output)(ifp, m0, sintosa(&dst), NULL); 5942 } else 5943 m_freem(m0); 5944 } 5945 5946 if (error == 0) 5947 KMOD_IPSTAT_INC(ips_fragmented); 5948 5949 done: 5950 if (r->rt != PF_DUPTO) 5951 *m = NULL; 5952 return; 5953 5954 bad_locked: 5955 if (s) 5956 PF_STATE_UNLOCK(s); 5957 bad: 5958 m_freem(m0); 5959 goto done; 5960 } 5961 #endif /* INET */ 5962 5963 #ifdef INET6 5964 static void 5965 pf_route6(struct mbuf **m, struct pf_krule *r, int dir, struct ifnet *oifp, 5966 struct pf_kstate *s, struct pf_pdesc *pd, struct inpcb *inp) 5967 { 5968 struct mbuf *m0; 5969 struct sockaddr_in6 dst; 5970 struct ip6_hdr *ip6; 5971 struct ifnet *ifp = NULL; 5972 struct pf_addr naddr; 5973 struct pf_ksrc_node *sn = NULL; 5974 5975 KASSERT(m && *m && r && oifp, ("%s: invalid parameters", __func__)); 5976 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: invalid direction", 5977 __func__)); 5978 5979 if ((pd->pf_mtag == NULL && 5980 ((pd->pf_mtag = pf_get_mtag(*m)) == NULL)) || 5981 pd->pf_mtag->routed++ > 3) { 5982 m0 = *m; 5983 *m = NULL; 5984 goto bad_locked; 5985 } 5986 5987 if (r->rt == PF_DUPTO) { 5988 if ((pd->pf_mtag->flags & PF_DUPLICATED)) { 5989 if (s == NULL) { 5990 ifp = r->rpool.cur->kif ? 5991 r->rpool.cur->kif->pfik_ifp : NULL; 5992 } else { 5993 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 5994 PF_STATE_UNLOCK(s); 5995 } 5996 if (ifp == oifp) { 5997 /* When the 2nd interface is not skipped */ 5998 return; 5999 } else { 6000 m0 = *m; 6001 *m = NULL; 6002 goto bad; 6003 } 6004 } else { 6005 pd->pf_mtag->flags |= PF_DUPLICATED; 6006 if (((m0 = m_dup(*m, M_NOWAIT)) == NULL)) { 6007 if (s) 6008 PF_STATE_UNLOCK(s); 6009 return; 6010 } 6011 } 6012 } else { 6013 if ((r->rt == PF_REPLYTO) == (r->direction == dir)) { 6014 if (s) 6015 PF_STATE_UNLOCK(s); 6016 return; 6017 } 6018 m0 = *m; 6019 } 6020 6021 ip6 = mtod(m0, struct ip6_hdr *); 6022 6023 bzero(&dst, sizeof(dst)); 6024 dst.sin6_family = AF_INET6; 6025 dst.sin6_len = sizeof(dst); 6026 dst.sin6_addr = ip6->ip6_dst; 6027 6028 bzero(&naddr, sizeof(naddr)); 6029 6030 if (TAILQ_EMPTY(&r->rpool.list)) { 6031 DPFPRINTF(PF_DEBUG_URGENT, 6032 ("%s: TAILQ_EMPTY(&r->rpool.list)\n", __func__)); 6033 goto bad_locked; 6034 } 6035 if (s == NULL) { 6036 pf_map_addr(AF_INET6, r, (struct pf_addr *)&ip6->ip6_src, 6037 &naddr, NULL, &sn); 6038 if (!PF_AZERO(&naddr, AF_INET6)) 6039 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 6040 &naddr, AF_INET6); 6041 ifp = r->rpool.cur->kif ? r->rpool.cur->kif->pfik_ifp : NULL; 6042 } else { 6043 if (!PF_AZERO(&s->rt_addr, AF_INET6)) 6044 PF_ACPY((struct pf_addr *)&dst.sin6_addr, 6045 &s->rt_addr, AF_INET6); 6046 ifp = s->rt_kif ? s->rt_kif->pfik_ifp : NULL; 6047 } 6048 6049 if (s) 6050 PF_STATE_UNLOCK(s); 6051 6052 if (ifp == NULL) 6053 goto bad; 6054 6055 if (dir == PF_IN) { 6056 if (pf_test6(PF_OUT, PFIL_FWD, ifp, &m0, inp) != PF_PASS) 6057 goto bad; 6058 else if (m0 == NULL) 6059 goto done; 6060 if (m0->m_len < sizeof(struct ip6_hdr)) { 6061 DPFPRINTF(PF_DEBUG_URGENT, 6062 ("%s: m0->m_len < sizeof(struct ip6_hdr)\n", 6063 __func__)); 6064 goto bad; 6065 } 6066 ip6 = mtod(m0, struct ip6_hdr *); 6067 } 6068 6069 if (ifp->if_flags & IFF_LOOPBACK) 6070 m0->m_flags |= M_SKIP_FIREWALL; 6071 6072 if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6 & 6073 ~ifp->if_hwassist) { 6074 uint32_t plen = m0->m_pkthdr.len - sizeof(*ip6); 6075 m0 = mb_unmapped_to_ext(m0); 6076 if (m0 == NULL) 6077 goto done; 6078 in6_delayed_cksum(m0, plen, sizeof(struct ip6_hdr)); 6079 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 6080 } 6081 6082 /* 6083 * If the packet is too large for the outgoing interface, 6084 * send back an icmp6 error. 6085 */ 6086 if (IN6_IS_SCOPE_EMBED(&dst.sin6_addr)) 6087 dst.sin6_addr.s6_addr16[1] = htons(ifp->if_index); 6088 if ((u_long)m0->m_pkthdr.len <= ifp->if_mtu) 6089 nd6_output_ifp(ifp, ifp, m0, &dst, NULL); 6090 else { 6091 in6_ifstat_inc(ifp, ifs6_in_toobig); 6092 if (r->rt != PF_DUPTO) 6093 icmp6_error(m0, ICMP6_PACKET_TOO_BIG, 0, ifp->if_mtu); 6094 else 6095 goto bad; 6096 } 6097 6098 done: 6099 if (r->rt != PF_DUPTO) 6100 *m = NULL; 6101 return; 6102 6103 bad_locked: 6104 if (s) 6105 PF_STATE_UNLOCK(s); 6106 bad: 6107 m_freem(m0); 6108 goto done; 6109 } 6110 #endif /* INET6 */ 6111 6112 /* 6113 * FreeBSD supports cksum offloads for the following drivers. 6114 * em(4), fxp(4), lge(4), nge(4), re(4), ti(4), txp(4), xl(4) 6115 * 6116 * CSUM_DATA_VALID | CSUM_PSEUDO_HDR : 6117 * network driver performed cksum including pseudo header, need to verify 6118 * csum_data 6119 * CSUM_DATA_VALID : 6120 * network driver performed cksum, needs to additional pseudo header 6121 * cksum computation with partial csum_data(i.e. lack of H/W support for 6122 * pseudo header, for instance sk(4) and possibly gem(4)) 6123 * 6124 * After validating the cksum of packet, set both flag CSUM_DATA_VALID and 6125 * CSUM_PSEUDO_HDR in order to avoid recomputation of the cksum in upper 6126 * TCP/UDP layer. 6127 * Also, set csum_data to 0xffff to force cksum validation. 6128 */ 6129 static int 6130 pf_check_proto_cksum(struct mbuf *m, int off, int len, u_int8_t p, sa_family_t af) 6131 { 6132 u_int16_t sum = 0; 6133 int hw_assist = 0; 6134 struct ip *ip; 6135 6136 if (off < sizeof(struct ip) || len < sizeof(struct udphdr)) 6137 return (1); 6138 if (m->m_pkthdr.len < off + len) 6139 return (1); 6140 6141 switch (p) { 6142 case IPPROTO_TCP: 6143 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 6144 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 6145 sum = m->m_pkthdr.csum_data; 6146 } else { 6147 ip = mtod(m, struct ip *); 6148 sum = in_pseudo(ip->ip_src.s_addr, 6149 ip->ip_dst.s_addr, htonl((u_short)len + 6150 m->m_pkthdr.csum_data + IPPROTO_TCP)); 6151 } 6152 sum ^= 0xffff; 6153 ++hw_assist; 6154 } 6155 break; 6156 case IPPROTO_UDP: 6157 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 6158 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) { 6159 sum = m->m_pkthdr.csum_data; 6160 } else { 6161 ip = mtod(m, struct ip *); 6162 sum = in_pseudo(ip->ip_src.s_addr, 6163 ip->ip_dst.s_addr, htonl((u_short)len + 6164 m->m_pkthdr.csum_data + IPPROTO_UDP)); 6165 } 6166 sum ^= 0xffff; 6167 ++hw_assist; 6168 } 6169 break; 6170 case IPPROTO_ICMP: 6171 #ifdef INET6 6172 case IPPROTO_ICMPV6: 6173 #endif /* INET6 */ 6174 break; 6175 default: 6176 return (1); 6177 } 6178 6179 if (!hw_assist) { 6180 switch (af) { 6181 case AF_INET: 6182 if (p == IPPROTO_ICMP) { 6183 if (m->m_len < off) 6184 return (1); 6185 m->m_data += off; 6186 m->m_len -= off; 6187 sum = in_cksum(m, len); 6188 m->m_data -= off; 6189 m->m_len += off; 6190 } else { 6191 if (m->m_len < sizeof(struct ip)) 6192 return (1); 6193 sum = in4_cksum(m, p, off, len); 6194 } 6195 break; 6196 #ifdef INET6 6197 case AF_INET6: 6198 if (m->m_len < sizeof(struct ip6_hdr)) 6199 return (1); 6200 sum = in6_cksum(m, p, off, len); 6201 break; 6202 #endif /* INET6 */ 6203 default: 6204 return (1); 6205 } 6206 } 6207 if (sum) { 6208 switch (p) { 6209 case IPPROTO_TCP: 6210 { 6211 KMOD_TCPSTAT_INC(tcps_rcvbadsum); 6212 break; 6213 } 6214 case IPPROTO_UDP: 6215 { 6216 KMOD_UDPSTAT_INC(udps_badsum); 6217 break; 6218 } 6219 #ifdef INET 6220 case IPPROTO_ICMP: 6221 { 6222 KMOD_ICMPSTAT_INC(icps_checksum); 6223 break; 6224 } 6225 #endif 6226 #ifdef INET6 6227 case IPPROTO_ICMPV6: 6228 { 6229 KMOD_ICMP6STAT_INC(icp6s_checksum); 6230 break; 6231 } 6232 #endif /* INET6 */ 6233 } 6234 return (1); 6235 } else { 6236 if (p == IPPROTO_TCP || p == IPPROTO_UDP) { 6237 m->m_pkthdr.csum_flags |= 6238 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); 6239 m->m_pkthdr.csum_data = 0xffff; 6240 } 6241 } 6242 return (0); 6243 } 6244 6245 static bool 6246 pf_pdesc_to_dnflow(int dir, const struct pf_pdesc *pd, 6247 const struct pf_krule *r, const struct pf_kstate *s, 6248 struct ip_fw_args *dnflow) 6249 { 6250 int dndir = r->direction; 6251 6252 if (s && dndir == PF_INOUT) 6253 dndir = s->direction; 6254 6255 memset(dnflow, 0, sizeof(*dnflow)); 6256 6257 if (pd->dport != NULL) 6258 dnflow->f_id.dst_port = ntohs(*pd->dport); 6259 if (pd->sport != NULL) 6260 dnflow->f_id.src_port = ntohs(*pd->sport); 6261 6262 if (dir == PF_IN) 6263 dnflow->flags |= IPFW_ARGS_IN; 6264 else 6265 dnflow->flags |= IPFW_ARGS_OUT; 6266 6267 if (dir != dndir && pd->act.dnrpipe) { 6268 dnflow->rule.info = pd->act.dnrpipe; 6269 } 6270 else if (dir == dndir) { 6271 dnflow->rule.info = pd->act.dnpipe; 6272 } 6273 else { 6274 return (false); 6275 } 6276 6277 dnflow->rule.info |= IPFW_IS_DUMMYNET; 6278 if (r->free_flags & PFRULE_DN_IS_PIPE) 6279 dnflow->rule.info |= IPFW_IS_PIPE; 6280 6281 dnflow->f_id.proto = pd->proto; 6282 dnflow->f_id.extra = dnflow->rule.info; 6283 switch (pd->af) { 6284 case AF_INET: 6285 dnflow->f_id.addr_type = 4; 6286 dnflow->f_id.src_ip = ntohl(pd->src->v4.s_addr); 6287 dnflow->f_id.dst_ip = ntohl(pd->dst->v4.s_addr); 6288 break; 6289 case AF_INET6: 6290 dnflow->flags |= IPFW_ARGS_IP6; 6291 dnflow->f_id.addr_type = 6; 6292 dnflow->f_id.src_ip6 = pd->src->v6; 6293 dnflow->f_id.dst_ip6 = pd->dst->v6; 6294 break; 6295 default: 6296 panic("Invalid AF"); 6297 break; 6298 } 6299 6300 return (true); 6301 } 6302 6303 #ifdef INET 6304 int 6305 pf_test(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) 6306 { 6307 struct pfi_kkif *kif; 6308 u_short action, reason = 0, log = 0; 6309 struct mbuf *m = *m0; 6310 struct ip *h = NULL; 6311 struct m_tag *ipfwtag; 6312 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 6313 struct pf_kstate *s = NULL; 6314 struct pf_kruleset *ruleset = NULL; 6315 struct pf_pdesc pd; 6316 int off, dirndx, pqid = 0; 6317 6318 PF_RULES_RLOCK_TRACKER; 6319 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 6320 M_ASSERTPKTHDR(m); 6321 6322 if (!V_pf_status.running) 6323 return (PF_PASS); 6324 6325 memset(&pd, 0, sizeof(pd)); 6326 6327 kif = (struct pfi_kkif *)ifp->if_pf_kif; 6328 6329 if (kif == NULL) { 6330 DPFPRINTF(PF_DEBUG_URGENT, 6331 ("pf_test: kif == NULL, if_xname %s\n", ifp->if_xname)); 6332 return (PF_DROP); 6333 } 6334 if (kif->pfik_flags & PFI_IFLAG_SKIP) 6335 return (PF_PASS); 6336 6337 if (m->m_flags & M_SKIP_FIREWALL) 6338 return (PF_PASS); 6339 6340 pd.pf_mtag = pf_find_mtag(m); 6341 6342 PF_RULES_RLOCK(); 6343 6344 if ((__predict_false(ip_divert_ptr != NULL) || ip_dn_io_ptr != NULL) && 6345 ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) { 6346 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1); 6347 if ((rr->info & IPFW_IS_DIVERT && rr->rulenum == 0) || 6348 (rr->info & IPFW_IS_DUMMYNET)) { 6349 if (pd.pf_mtag == NULL && 6350 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 6351 action = PF_DROP; 6352 goto done; 6353 } 6354 pd.pf_mtag->flags |= PF_PACKET_LOOPED; 6355 m_tag_delete(m, ipfwtag); 6356 } 6357 if (pd.pf_mtag && pd.pf_mtag->flags & PF_FASTFWD_OURS_PRESENT) { 6358 m->m_flags |= M_FASTFWD_OURS; 6359 pd.pf_mtag->flags &= ~PF_FASTFWD_OURS_PRESENT; 6360 } 6361 } else if (pf_normalize_ip(m0, dir, kif, &reason, &pd) != PF_PASS) { 6362 /* We do IP header normalization and packet reassembly here */ 6363 action = PF_DROP; 6364 goto done; 6365 } 6366 m = *m0; /* pf_normalize messes with m0 */ 6367 h = mtod(m, struct ip *); 6368 6369 off = h->ip_hl << 2; 6370 if (off < (int)sizeof(struct ip)) { 6371 action = PF_DROP; 6372 REASON_SET(&reason, PFRES_SHORT); 6373 log = 1; 6374 goto done; 6375 } 6376 6377 pd.src = (struct pf_addr *)&h->ip_src; 6378 pd.dst = (struct pf_addr *)&h->ip_dst; 6379 pd.sport = pd.dport = NULL; 6380 pd.ip_sum = &h->ip_sum; 6381 pd.proto_sum = NULL; 6382 pd.proto = h->ip_p; 6383 pd.dir = dir; 6384 pd.sidx = (dir == PF_IN) ? 0 : 1; 6385 pd.didx = (dir == PF_IN) ? 1 : 0; 6386 pd.af = AF_INET; 6387 pd.tos = h->ip_tos & ~IPTOS_ECN_MASK; 6388 pd.tot_len = ntohs(h->ip_len); 6389 6390 /* handle fragments that didn't get reassembled by normalization */ 6391 if (h->ip_off & htons(IP_MF | IP_OFFMASK)) { 6392 action = pf_test_fragment(&r, dir, kif, m, h, 6393 &pd, &a, &ruleset); 6394 goto done; 6395 } 6396 6397 switch (h->ip_p) { 6398 case IPPROTO_TCP: { 6399 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 6400 &action, &reason, AF_INET)) { 6401 log = action != PF_PASS; 6402 goto done; 6403 } 6404 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 6405 6406 pd.sport = &pd.hdr.tcp.th_sport; 6407 pd.dport = &pd.hdr.tcp.th_dport; 6408 6409 /* Respond to SYN with a syncookie. */ 6410 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == TH_SYN && 6411 pd.dir == PF_IN && pf_synflood_check(&pd)) { 6412 pf_syncookie_send(m, off, &pd); 6413 action = PF_DROP; 6414 break; 6415 } 6416 6417 if ((pd.hdr.tcp.th_flags & TH_ACK) && pd.p_len == 0) 6418 pqid = 1; 6419 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 6420 if (action == PF_DROP) 6421 goto done; 6422 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 6423 &reason); 6424 if (action == PF_PASS) { 6425 if (V_pfsync_update_state_ptr != NULL) 6426 V_pfsync_update_state_ptr(s); 6427 r = s->rule.ptr; 6428 a = s->anchor.ptr; 6429 log = s->log; 6430 } else if (s == NULL) { 6431 /* Validate remote SYN|ACK, re-create original SYN if 6432 * valid. */ 6433 if ((pd.hdr.tcp.th_flags & (TH_SYN|TH_ACK|TH_RST)) == 6434 TH_ACK && pf_syncookie_validate(&pd) && 6435 pd.dir == PF_IN) { 6436 struct mbuf *msyn; 6437 6438 msyn = pf_syncookie_recreate_syn(h->ip_ttl, 6439 off,&pd); 6440 if (msyn == NULL) { 6441 action = PF_DROP; 6442 break; 6443 } 6444 6445 action = pf_test(dir, pflags, ifp, &msyn, inp); 6446 m_freem(msyn); 6447 6448 if (action == PF_PASS) { 6449 action = pf_test_state_tcp(&s, dir, 6450 kif, m, off, h, &pd, &reason); 6451 if (action != PF_PASS || s == NULL) { 6452 action = PF_DROP; 6453 break; 6454 } 6455 6456 s->src.seqhi = ntohl(pd.hdr.tcp.th_ack) 6457 - 1; 6458 s->src.seqlo = ntohl(pd.hdr.tcp.th_seq) 6459 - 1; 6460 pf_set_protostate(s, PF_PEER_SRC, 6461 PF_TCPS_PROXY_DST); 6462 6463 action = pf_synproxy(&pd, &s, &reason); 6464 if (action != PF_PASS) 6465 break; 6466 } 6467 break; 6468 } 6469 else { 6470 action = pf_test_rule(&r, &s, dir, kif, m, off, 6471 &pd, &a, &ruleset, inp); 6472 } 6473 } 6474 break; 6475 } 6476 6477 case IPPROTO_UDP: { 6478 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 6479 &action, &reason, AF_INET)) { 6480 log = action != PF_PASS; 6481 goto done; 6482 } 6483 pd.sport = &pd.hdr.udp.uh_sport; 6484 pd.dport = &pd.hdr.udp.uh_dport; 6485 if (pd.hdr.udp.uh_dport == 0 || 6486 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 6487 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 6488 action = PF_DROP; 6489 REASON_SET(&reason, PFRES_SHORT); 6490 goto done; 6491 } 6492 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 6493 if (action == PF_PASS) { 6494 if (V_pfsync_update_state_ptr != NULL) 6495 V_pfsync_update_state_ptr(s); 6496 r = s->rule.ptr; 6497 a = s->anchor.ptr; 6498 log = s->log; 6499 } else if (s == NULL) 6500 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6501 &a, &ruleset, inp); 6502 break; 6503 } 6504 6505 case IPPROTO_ICMP: { 6506 if (!pf_pull_hdr(m, off, &pd.hdr.icmp, ICMP_MINLEN, 6507 &action, &reason, AF_INET)) { 6508 log = action != PF_PASS; 6509 goto done; 6510 } 6511 action = pf_test_state_icmp(&s, dir, kif, m, off, h, &pd, 6512 &reason); 6513 if (action == PF_PASS) { 6514 if (V_pfsync_update_state_ptr != NULL) 6515 V_pfsync_update_state_ptr(s); 6516 r = s->rule.ptr; 6517 a = s->anchor.ptr; 6518 log = s->log; 6519 } else if (s == NULL) 6520 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6521 &a, &ruleset, inp); 6522 break; 6523 } 6524 6525 #ifdef INET6 6526 case IPPROTO_ICMPV6: { 6527 action = PF_DROP; 6528 DPFPRINTF(PF_DEBUG_MISC, 6529 ("pf: dropping IPv4 packet with ICMPv6 payload\n")); 6530 goto done; 6531 } 6532 #endif 6533 6534 default: 6535 action = pf_test_state_other(&s, dir, kif, m, &pd); 6536 if (action == PF_PASS) { 6537 if (V_pfsync_update_state_ptr != NULL) 6538 V_pfsync_update_state_ptr(s); 6539 r = s->rule.ptr; 6540 a = s->anchor.ptr; 6541 log = s->log; 6542 } else if (s == NULL) 6543 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6544 &a, &ruleset, inp); 6545 break; 6546 } 6547 6548 done: 6549 PF_RULES_RUNLOCK(); 6550 if (action == PF_PASS && h->ip_hl > 5 && 6551 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 6552 action = PF_DROP; 6553 REASON_SET(&reason, PFRES_IPOPTIONS); 6554 log = r->log; 6555 DPFPRINTF(PF_DEBUG_MISC, 6556 ("pf: dropping packet with ip options\n")); 6557 } 6558 6559 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { 6560 action = PF_DROP; 6561 REASON_SET(&reason, PFRES_MEMORY); 6562 } 6563 if (r->rtableid >= 0) 6564 M_SETFIB(m, r->rtableid); 6565 6566 if (r->scrub_flags & PFSTATE_SETPRIO) { 6567 if (pd.tos & IPTOS_LOWDELAY) 6568 pqid = 1; 6569 if (vlan_set_pcp(m, r->set_prio[pqid])) { 6570 action = PF_DROP; 6571 REASON_SET(&reason, PFRES_MEMORY); 6572 log = 1; 6573 DPFPRINTF(PF_DEBUG_MISC, 6574 ("pf: failed to allocate 802.1q mtag\n")); 6575 } 6576 } 6577 6578 #ifdef ALTQ 6579 if (s && s->qid) { 6580 pd.act.pqid = s->pqid; 6581 pd.act.qid = s->qid; 6582 } else if (r->qid) { 6583 pd.act.pqid = r->pqid; 6584 pd.act.qid = r->qid; 6585 } 6586 if (action == PF_PASS && pd.act.qid) { 6587 if (pd.pf_mtag == NULL && 6588 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 6589 action = PF_DROP; 6590 REASON_SET(&reason, PFRES_MEMORY); 6591 } else { 6592 if (s != NULL) 6593 pd.pf_mtag->qid_hash = pf_state_hash(s); 6594 if (pqid || (pd.tos & IPTOS_LOWDELAY)) 6595 pd.pf_mtag->qid = pd.act.pqid; 6596 else 6597 pd.pf_mtag->qid = pd.act.qid; 6598 /* Add hints for ecn. */ 6599 pd.pf_mtag->hdr = h; 6600 } 6601 } 6602 #endif /* ALTQ */ 6603 6604 if (s && (s->dnpipe || s->dnrpipe)) { 6605 pd.act.dnpipe = s->dnpipe; 6606 pd.act.dnrpipe = s->dnrpipe; 6607 pd.act.flags = s->state_flags; 6608 } else if (r->dnpipe || r->dnrpipe) { 6609 pd.act.dnpipe = r->dnpipe; 6610 pd.act.dnrpipe = r->dnrpipe; 6611 pd.act.flags = r->free_flags; 6612 } 6613 if ((pd.act.dnpipe || pd.act.dnrpipe) && !PACKET_LOOPED(&pd)) { 6614 if (ip_dn_io_ptr == NULL) { 6615 action = PF_DROP; 6616 REASON_SET(&reason, PFRES_MEMORY); 6617 } else { 6618 struct ip_fw_args dnflow; 6619 6620 if (pd.pf_mtag == NULL && 6621 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 6622 action = PF_DROP; 6623 REASON_SET(&reason, PFRES_MEMORY); 6624 if (s) 6625 PF_STATE_UNLOCK(s); 6626 return (action); 6627 } 6628 6629 if (pf_pdesc_to_dnflow(dir, &pd, r, s, &dnflow)) { 6630 ip_dn_io_ptr(m0, &dnflow); 6631 6632 if (*m0 == NULL) { 6633 if (s) 6634 PF_STATE_UNLOCK(s); 6635 return (action); 6636 } else { 6637 /* This is dummynet fast io processing */ 6638 m_tag_delete(*m0, m_tag_first(*m0)); 6639 pd.pf_mtag->flags &= ~PF_PACKET_LOOPED; 6640 } 6641 } 6642 } 6643 } 6644 6645 /* 6646 * connections redirected to loopback should not match sockets 6647 * bound specifically to loopback due to security implications, 6648 * see tcp_input() and in_pcblookup_listen(). 6649 */ 6650 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 6651 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 6652 (s->nat_rule.ptr->action == PF_RDR || 6653 s->nat_rule.ptr->action == PF_BINAT) && 6654 IN_LOOPBACK(ntohl(pd.dst->v4.s_addr))) 6655 m->m_flags |= M_SKIP_FIREWALL; 6656 6657 if (__predict_false(ip_divert_ptr != NULL) && action == PF_PASS && 6658 r->divert.port && !PACKET_LOOPED(&pd)) { 6659 ipfwtag = m_tag_alloc(MTAG_IPFW_RULE, 0, 6660 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); 6661 if (ipfwtag != NULL) { 6662 ((struct ipfw_rule_ref *)(ipfwtag+1))->info = 6663 ntohs(r->divert.port); 6664 ((struct ipfw_rule_ref *)(ipfwtag+1))->rulenum = dir; 6665 6666 if (s) 6667 PF_STATE_UNLOCK(s); 6668 6669 m_tag_prepend(m, ipfwtag); 6670 if (m->m_flags & M_FASTFWD_OURS) { 6671 if (pd.pf_mtag == NULL && 6672 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 6673 action = PF_DROP; 6674 REASON_SET(&reason, PFRES_MEMORY); 6675 log = 1; 6676 DPFPRINTF(PF_DEBUG_MISC, 6677 ("pf: failed to allocate tag\n")); 6678 } else { 6679 pd.pf_mtag->flags |= 6680 PF_FASTFWD_OURS_PRESENT; 6681 m->m_flags &= ~M_FASTFWD_OURS; 6682 } 6683 } 6684 ip_divert_ptr(*m0, dir == PF_IN); 6685 *m0 = NULL; 6686 6687 return (action); 6688 } else { 6689 /* XXX: ipfw has the same behaviour! */ 6690 action = PF_DROP; 6691 REASON_SET(&reason, PFRES_MEMORY); 6692 log = 1; 6693 DPFPRINTF(PF_DEBUG_MISC, 6694 ("pf: failed to allocate divert tag\n")); 6695 } 6696 } 6697 6698 if (log) { 6699 struct pf_krule *lr; 6700 6701 if (s != NULL && s->nat_rule.ptr != NULL && 6702 s->nat_rule.ptr->log & PF_LOG_ALL) 6703 lr = s->nat_rule.ptr; 6704 else 6705 lr = r; 6706 PFLOG_PACKET(kif, m, AF_INET, dir, reason, lr, a, ruleset, &pd, 6707 (s == NULL)); 6708 } 6709 6710 pf_counter_u64_critical_enter(); 6711 pf_counter_u64_add_protected(&kif->pfik_bytes[0][dir == PF_OUT][action != PF_PASS], 6712 pd.tot_len); 6713 pf_counter_u64_add_protected(&kif->pfik_packets[0][dir == PF_OUT][action != PF_PASS], 6714 1); 6715 6716 if (action == PF_PASS || r->action == PF_DROP) { 6717 dirndx = (dir == PF_OUT); 6718 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 6719 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 6720 if (a != NULL) { 6721 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 6722 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 6723 } 6724 if (s != NULL) { 6725 if (s->nat_rule.ptr != NULL) { 6726 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 6727 1); 6728 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 6729 pd.tot_len); 6730 } 6731 if (s->src_node != NULL) { 6732 counter_u64_add(s->src_node->packets[dirndx], 6733 1); 6734 counter_u64_add(s->src_node->bytes[dirndx], 6735 pd.tot_len); 6736 } 6737 if (s->nat_src_node != NULL) { 6738 counter_u64_add(s->nat_src_node->packets[dirndx], 6739 1); 6740 counter_u64_add(s->nat_src_node->bytes[dirndx], 6741 pd.tot_len); 6742 } 6743 dirndx = (dir == s->direction) ? 0 : 1; 6744 s->packets[dirndx]++; 6745 s->bytes[dirndx] += pd.tot_len; 6746 } 6747 tr = r; 6748 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 6749 if (nr != NULL && r == &V_pf_default_rule) 6750 tr = nr; 6751 if (tr->src.addr.type == PF_ADDR_TABLE) 6752 pfr_update_stats(tr->src.addr.p.tbl, 6753 (s == NULL) ? pd.src : 6754 &s->key[(s->direction == PF_IN)]-> 6755 addr[(s->direction == PF_OUT)], 6756 pd.af, pd.tot_len, dir == PF_OUT, 6757 r->action == PF_PASS, tr->src.neg); 6758 if (tr->dst.addr.type == PF_ADDR_TABLE) 6759 pfr_update_stats(tr->dst.addr.p.tbl, 6760 (s == NULL) ? pd.dst : 6761 &s->key[(s->direction == PF_IN)]-> 6762 addr[(s->direction == PF_IN)], 6763 pd.af, pd.tot_len, dir == PF_OUT, 6764 r->action == PF_PASS, tr->dst.neg); 6765 } 6766 pf_counter_u64_critical_exit(); 6767 6768 switch (action) { 6769 case PF_SYNPROXY_DROP: 6770 m_freem(*m0); 6771 case PF_DEFER: 6772 *m0 = NULL; 6773 action = PF_PASS; 6774 break; 6775 case PF_DROP: 6776 m_freem(*m0); 6777 *m0 = NULL; 6778 break; 6779 default: 6780 /* pf_route() returns unlocked. */ 6781 if (r->rt) { 6782 pf_route(m0, r, dir, kif->pfik_ifp, s, &pd, inp); 6783 return (action); 6784 } 6785 break; 6786 } 6787 6788 SDT_PROBE4(pf, ip, test, done, action, reason, r, s); 6789 6790 if (s) 6791 PF_STATE_UNLOCK(s); 6792 6793 return (action); 6794 } 6795 #endif /* INET */ 6796 6797 #ifdef INET6 6798 int 6799 pf_test6(int dir, int pflags, struct ifnet *ifp, struct mbuf **m0, struct inpcb *inp) 6800 { 6801 struct pfi_kkif *kif; 6802 u_short action, reason = 0, log = 0; 6803 struct mbuf *m = *m0, *n = NULL; 6804 struct m_tag *mtag; 6805 struct m_tag *ipfwtag; 6806 struct ip6_hdr *h = NULL; 6807 struct pf_krule *a = NULL, *r = &V_pf_default_rule, *tr, *nr; 6808 struct pf_kstate *s = NULL; 6809 struct pf_kruleset *ruleset = NULL; 6810 struct pf_pdesc pd; 6811 int off, terminal = 0, dirndx, rh_cnt = 0, pqid = 0; 6812 6813 PF_RULES_RLOCK_TRACKER; 6814 KASSERT(dir == PF_IN || dir == PF_OUT, ("%s: bad direction %d\n", __func__, dir)); 6815 M_ASSERTPKTHDR(m); 6816 6817 if (!V_pf_status.running) 6818 return (PF_PASS); 6819 6820 memset(&pd, 0, sizeof(pd)); 6821 pd.pf_mtag = pf_find_mtag(m); 6822 6823 if (pd.pf_mtag && pd.pf_mtag->flags & PF_TAG_GENERATED) 6824 return (PF_PASS); 6825 6826 kif = (struct pfi_kkif *)ifp->if_pf_kif; 6827 if (kif == NULL) { 6828 DPFPRINTF(PF_DEBUG_URGENT, 6829 ("pf_test6: kif == NULL, if_xname %s\n", ifp->if_xname)); 6830 return (PF_DROP); 6831 } 6832 if (kif->pfik_flags & PFI_IFLAG_SKIP) 6833 return (PF_PASS); 6834 6835 if (m->m_flags & M_SKIP_FIREWALL) 6836 return (PF_PASS); 6837 6838 PF_RULES_RLOCK(); 6839 6840 /* We do IP header normalization and packet reassembly here */ 6841 if (ip_dn_io_ptr != NULL && 6842 ((ipfwtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL)) != NULL)) { 6843 struct ipfw_rule_ref *rr = (struct ipfw_rule_ref *)(ipfwtag+1); 6844 if (rr->info & IPFW_IS_DUMMYNET) { 6845 if (pd.pf_mtag == NULL && 6846 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 6847 action = PF_DROP; 6848 goto done; 6849 } 6850 pd.pf_mtag->flags |= PF_PACKET_LOOPED; 6851 m_tag_delete(m, ipfwtag); 6852 } 6853 } else if (pf_normalize_ip6(m0, dir, kif, &reason, &pd) != PF_PASS) { 6854 action = PF_DROP; 6855 goto done; 6856 } 6857 m = *m0; /* pf_normalize messes with m0 */ 6858 h = mtod(m, struct ip6_hdr *); 6859 6860 /* 6861 * we do not support jumbogram. if we keep going, zero ip6_plen 6862 * will do something bad, so drop the packet for now. 6863 */ 6864 if (htons(h->ip6_plen) == 0) { 6865 action = PF_DROP; 6866 REASON_SET(&reason, PFRES_NORM); /*XXX*/ 6867 goto done; 6868 } 6869 6870 pd.src = (struct pf_addr *)&h->ip6_src; 6871 pd.dst = (struct pf_addr *)&h->ip6_dst; 6872 pd.sport = pd.dport = NULL; 6873 pd.ip_sum = NULL; 6874 pd.proto_sum = NULL; 6875 pd.dir = dir; 6876 pd.sidx = (dir == PF_IN) ? 0 : 1; 6877 pd.didx = (dir == PF_IN) ? 1 : 0; 6878 pd.af = AF_INET6; 6879 pd.tos = IPV6_DSCP(h); 6880 pd.tot_len = ntohs(h->ip6_plen) + sizeof(struct ip6_hdr); 6881 6882 off = ((caddr_t)h - m->m_data) + sizeof(struct ip6_hdr); 6883 pd.proto = h->ip6_nxt; 6884 do { 6885 switch (pd.proto) { 6886 case IPPROTO_FRAGMENT: 6887 action = pf_test_fragment(&r, dir, kif, m, h, 6888 &pd, &a, &ruleset); 6889 if (action == PF_DROP) 6890 REASON_SET(&reason, PFRES_FRAG); 6891 goto done; 6892 case IPPROTO_ROUTING: { 6893 struct ip6_rthdr rthdr; 6894 6895 if (rh_cnt++) { 6896 DPFPRINTF(PF_DEBUG_MISC, 6897 ("pf: IPv6 more than one rthdr\n")); 6898 action = PF_DROP; 6899 REASON_SET(&reason, PFRES_IPOPTIONS); 6900 log = 1; 6901 goto done; 6902 } 6903 if (!pf_pull_hdr(m, off, &rthdr, sizeof(rthdr), NULL, 6904 &reason, pd.af)) { 6905 DPFPRINTF(PF_DEBUG_MISC, 6906 ("pf: IPv6 short rthdr\n")); 6907 action = PF_DROP; 6908 REASON_SET(&reason, PFRES_SHORT); 6909 log = 1; 6910 goto done; 6911 } 6912 if (rthdr.ip6r_type == IPV6_RTHDR_TYPE_0) { 6913 DPFPRINTF(PF_DEBUG_MISC, 6914 ("pf: IPv6 rthdr0\n")); 6915 action = PF_DROP; 6916 REASON_SET(&reason, PFRES_IPOPTIONS); 6917 log = 1; 6918 goto done; 6919 } 6920 /* FALLTHROUGH */ 6921 } 6922 case IPPROTO_AH: 6923 case IPPROTO_HOPOPTS: 6924 case IPPROTO_DSTOPTS: { 6925 /* get next header and header length */ 6926 struct ip6_ext opt6; 6927 6928 if (!pf_pull_hdr(m, off, &opt6, sizeof(opt6), 6929 NULL, &reason, pd.af)) { 6930 DPFPRINTF(PF_DEBUG_MISC, 6931 ("pf: IPv6 short opt\n")); 6932 action = PF_DROP; 6933 log = 1; 6934 goto done; 6935 } 6936 if (pd.proto == IPPROTO_AH) 6937 off += (opt6.ip6e_len + 2) * 4; 6938 else 6939 off += (opt6.ip6e_len + 1) * 8; 6940 pd.proto = opt6.ip6e_nxt; 6941 /* goto the next header */ 6942 break; 6943 } 6944 default: 6945 terminal++; 6946 break; 6947 } 6948 } while (!terminal); 6949 6950 /* if there's no routing header, use unmodified mbuf for checksumming */ 6951 if (!n) 6952 n = m; 6953 6954 switch (pd.proto) { 6955 case IPPROTO_TCP: { 6956 if (!pf_pull_hdr(m, off, &pd.hdr.tcp, sizeof(pd.hdr.tcp), 6957 &action, &reason, AF_INET6)) { 6958 log = action != PF_PASS; 6959 goto done; 6960 } 6961 pd.p_len = pd.tot_len - off - (pd.hdr.tcp.th_off << 2); 6962 pd.sport = &pd.hdr.tcp.th_sport; 6963 pd.dport = &pd.hdr.tcp.th_dport; 6964 action = pf_normalize_tcp(dir, kif, m, 0, off, h, &pd); 6965 if (action == PF_DROP) 6966 goto done; 6967 action = pf_test_state_tcp(&s, dir, kif, m, off, h, &pd, 6968 &reason); 6969 if (action == PF_PASS) { 6970 if (V_pfsync_update_state_ptr != NULL) 6971 V_pfsync_update_state_ptr(s); 6972 r = s->rule.ptr; 6973 a = s->anchor.ptr; 6974 log = s->log; 6975 } else if (s == NULL) 6976 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 6977 &a, &ruleset, inp); 6978 break; 6979 } 6980 6981 case IPPROTO_UDP: { 6982 if (!pf_pull_hdr(m, off, &pd.hdr.udp, sizeof(pd.hdr.udp), 6983 &action, &reason, AF_INET6)) { 6984 log = action != PF_PASS; 6985 goto done; 6986 } 6987 pd.sport = &pd.hdr.udp.uh_sport; 6988 pd.dport = &pd.hdr.udp.uh_dport; 6989 if (pd.hdr.udp.uh_dport == 0 || 6990 ntohs(pd.hdr.udp.uh_ulen) > m->m_pkthdr.len - off || 6991 ntohs(pd.hdr.udp.uh_ulen) < sizeof(struct udphdr)) { 6992 action = PF_DROP; 6993 REASON_SET(&reason, PFRES_SHORT); 6994 goto done; 6995 } 6996 action = pf_test_state_udp(&s, dir, kif, m, off, h, &pd); 6997 if (action == PF_PASS) { 6998 if (V_pfsync_update_state_ptr != NULL) 6999 V_pfsync_update_state_ptr(s); 7000 r = s->rule.ptr; 7001 a = s->anchor.ptr; 7002 log = s->log; 7003 } else if (s == NULL) 7004 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7005 &a, &ruleset, inp); 7006 break; 7007 } 7008 7009 case IPPROTO_ICMP: { 7010 action = PF_DROP; 7011 DPFPRINTF(PF_DEBUG_MISC, 7012 ("pf: dropping IPv6 packet with ICMPv4 payload\n")); 7013 goto done; 7014 } 7015 7016 case IPPROTO_ICMPV6: { 7017 if (!pf_pull_hdr(m, off, &pd.hdr.icmp6, sizeof(pd.hdr.icmp6), 7018 &action, &reason, AF_INET6)) { 7019 log = action != PF_PASS; 7020 goto done; 7021 } 7022 action = pf_test_state_icmp(&s, dir, kif, 7023 m, off, h, &pd, &reason); 7024 if (action == PF_PASS) { 7025 if (V_pfsync_update_state_ptr != NULL) 7026 V_pfsync_update_state_ptr(s); 7027 r = s->rule.ptr; 7028 a = s->anchor.ptr; 7029 log = s->log; 7030 } else if (s == NULL) 7031 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7032 &a, &ruleset, inp); 7033 break; 7034 } 7035 7036 default: 7037 action = pf_test_state_other(&s, dir, kif, m, &pd); 7038 if (action == PF_PASS) { 7039 if (V_pfsync_update_state_ptr != NULL) 7040 V_pfsync_update_state_ptr(s); 7041 r = s->rule.ptr; 7042 a = s->anchor.ptr; 7043 log = s->log; 7044 } else if (s == NULL) 7045 action = pf_test_rule(&r, &s, dir, kif, m, off, &pd, 7046 &a, &ruleset, inp); 7047 break; 7048 } 7049 7050 done: 7051 PF_RULES_RUNLOCK(); 7052 if (n != m) { 7053 m_freem(n); 7054 n = NULL; 7055 } 7056 7057 /* handle dangerous IPv6 extension headers. */ 7058 if (action == PF_PASS && rh_cnt && 7059 !((s && s->state_flags & PFSTATE_ALLOWOPTS) || r->allow_opts)) { 7060 action = PF_DROP; 7061 REASON_SET(&reason, PFRES_IPOPTIONS); 7062 log = r->log; 7063 DPFPRINTF(PF_DEBUG_MISC, 7064 ("pf: dropping packet with dangerous v6 headers\n")); 7065 } 7066 7067 if (s && s->tag > 0 && pf_tag_packet(m, &pd, s->tag)) { 7068 action = PF_DROP; 7069 REASON_SET(&reason, PFRES_MEMORY); 7070 } 7071 if (r->rtableid >= 0) 7072 M_SETFIB(m, r->rtableid); 7073 7074 if (r->scrub_flags & PFSTATE_SETPRIO) { 7075 if (pd.tos & IPTOS_LOWDELAY) 7076 pqid = 1; 7077 if (vlan_set_pcp(m, r->set_prio[pqid])) { 7078 action = PF_DROP; 7079 REASON_SET(&reason, PFRES_MEMORY); 7080 log = 1; 7081 DPFPRINTF(PF_DEBUG_MISC, 7082 ("pf: failed to allocate 802.1q mtag\n")); 7083 } 7084 } 7085 7086 #ifdef ALTQ 7087 if (s && s->qid) { 7088 pd.act.pqid = s->pqid; 7089 pd.act.qid = s->qid; 7090 } else if (r->qid) { 7091 pd.act.pqid = r->pqid; 7092 pd.act.qid = r->qid; 7093 } 7094 if (action == PF_PASS && pd.act.qid) { 7095 if (pd.pf_mtag == NULL && 7096 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 7097 action = PF_DROP; 7098 REASON_SET(&reason, PFRES_MEMORY); 7099 } else { 7100 if (s != NULL) 7101 pd.pf_mtag->qid_hash = pf_state_hash(s); 7102 if (pd.tos & IPTOS_LOWDELAY) 7103 pd.pf_mtag->qid = pd.act.pqid; 7104 else 7105 pd.pf_mtag->qid = pd.act.qid; 7106 /* Add hints for ecn. */ 7107 pd.pf_mtag->hdr = h; 7108 } 7109 } 7110 #endif /* ALTQ */ 7111 7112 if (s && (s->dnpipe || s->dnrpipe)) { 7113 pd.act.dnpipe = s->dnpipe; 7114 pd.act.dnrpipe = s->dnrpipe; 7115 pd.act.flags = s->state_flags; 7116 } else { 7117 pd.act.dnpipe = r->dnpipe; 7118 pd.act.dnrpipe = r->dnrpipe; 7119 pd.act.flags = r->free_flags; 7120 } 7121 if ((pd.act.dnpipe || pd.act.dnrpipe) && !PACKET_LOOPED(&pd)) { 7122 if (ip_dn_io_ptr == NULL) { 7123 action = PF_DROP; 7124 REASON_SET(&reason, PFRES_MEMORY); 7125 } else { 7126 struct ip_fw_args dnflow; 7127 7128 if (pd.pf_mtag == NULL && 7129 ((pd.pf_mtag = pf_get_mtag(m)) == NULL)) { 7130 action = PF_DROP; 7131 REASON_SET(&reason, PFRES_MEMORY); 7132 if (s) 7133 PF_STATE_UNLOCK(s); 7134 return (action); 7135 } 7136 7137 if (pf_pdesc_to_dnflow(dir, &pd, r, s, &dnflow)) { 7138 ip_dn_io_ptr(m0, &dnflow); 7139 7140 if (*m0 == NULL) { 7141 if (s) 7142 PF_STATE_UNLOCK(s); 7143 return (action); 7144 } else { 7145 /* This is dummynet fast io processing */ 7146 m_tag_delete(*m0, m_tag_first(*m0)); 7147 pd.pf_mtag->flags &= ~PF_PACKET_LOOPED; 7148 } 7149 } 7150 } 7151 } 7152 7153 if (dir == PF_IN && action == PF_PASS && (pd.proto == IPPROTO_TCP || 7154 pd.proto == IPPROTO_UDP) && s != NULL && s->nat_rule.ptr != NULL && 7155 (s->nat_rule.ptr->action == PF_RDR || 7156 s->nat_rule.ptr->action == PF_BINAT) && 7157 IN6_IS_ADDR_LOOPBACK(&pd.dst->v6)) 7158 m->m_flags |= M_SKIP_FIREWALL; 7159 7160 /* XXX: Anybody working on it?! */ 7161 if (r->divert.port) 7162 printf("pf: divert(9) is not supported for IPv6\n"); 7163 7164 if (log) { 7165 struct pf_krule *lr; 7166 7167 if (s != NULL && s->nat_rule.ptr != NULL && 7168 s->nat_rule.ptr->log & PF_LOG_ALL) 7169 lr = s->nat_rule.ptr; 7170 else 7171 lr = r; 7172 PFLOG_PACKET(kif, m, AF_INET6, dir, reason, lr, a, ruleset, 7173 &pd, (s == NULL)); 7174 } 7175 7176 pf_counter_u64_critical_enter(); 7177 pf_counter_u64_add_protected(&kif->pfik_bytes[1][dir == PF_OUT][action != PF_PASS], 7178 pd.tot_len); 7179 pf_counter_u64_add_protected(&kif->pfik_packets[1][dir == PF_OUT][action != PF_PASS], 7180 1); 7181 7182 if (action == PF_PASS || r->action == PF_DROP) { 7183 dirndx = (dir == PF_OUT); 7184 pf_counter_u64_add_protected(&r->packets[dirndx], 1); 7185 pf_counter_u64_add_protected(&r->bytes[dirndx], pd.tot_len); 7186 if (a != NULL) { 7187 pf_counter_u64_add_protected(&a->packets[dirndx], 1); 7188 pf_counter_u64_add_protected(&a->bytes[dirndx], pd.tot_len); 7189 } 7190 if (s != NULL) { 7191 if (s->nat_rule.ptr != NULL) { 7192 pf_counter_u64_add_protected(&s->nat_rule.ptr->packets[dirndx], 7193 1); 7194 pf_counter_u64_add_protected(&s->nat_rule.ptr->bytes[dirndx], 7195 pd.tot_len); 7196 } 7197 if (s->src_node != NULL) { 7198 counter_u64_add(s->src_node->packets[dirndx], 7199 1); 7200 counter_u64_add(s->src_node->bytes[dirndx], 7201 pd.tot_len); 7202 } 7203 if (s->nat_src_node != NULL) { 7204 counter_u64_add(s->nat_src_node->packets[dirndx], 7205 1); 7206 counter_u64_add(s->nat_src_node->bytes[dirndx], 7207 pd.tot_len); 7208 } 7209 dirndx = (dir == s->direction) ? 0 : 1; 7210 s->packets[dirndx]++; 7211 s->bytes[dirndx] += pd.tot_len; 7212 } 7213 tr = r; 7214 nr = (s != NULL) ? s->nat_rule.ptr : pd.nat_rule; 7215 if (nr != NULL && r == &V_pf_default_rule) 7216 tr = nr; 7217 if (tr->src.addr.type == PF_ADDR_TABLE) 7218 pfr_update_stats(tr->src.addr.p.tbl, 7219 (s == NULL) ? pd.src : 7220 &s->key[(s->direction == PF_IN)]->addr[0], 7221 pd.af, pd.tot_len, dir == PF_OUT, 7222 r->action == PF_PASS, tr->src.neg); 7223 if (tr->dst.addr.type == PF_ADDR_TABLE) 7224 pfr_update_stats(tr->dst.addr.p.tbl, 7225 (s == NULL) ? pd.dst : 7226 &s->key[(s->direction == PF_IN)]->addr[1], 7227 pd.af, pd.tot_len, dir == PF_OUT, 7228 r->action == PF_PASS, tr->dst.neg); 7229 } 7230 pf_counter_u64_critical_exit(); 7231 7232 switch (action) { 7233 case PF_SYNPROXY_DROP: 7234 m_freem(*m0); 7235 case PF_DEFER: 7236 *m0 = NULL; 7237 action = PF_PASS; 7238 break; 7239 case PF_DROP: 7240 m_freem(*m0); 7241 *m0 = NULL; 7242 break; 7243 default: 7244 /* pf_route6() returns unlocked. */ 7245 if (r->rt) { 7246 pf_route6(m0, r, dir, kif->pfik_ifp, s, &pd, inp); 7247 return (action); 7248 } 7249 break; 7250 } 7251 7252 if (s) 7253 PF_STATE_UNLOCK(s); 7254 7255 /* If reassembled packet passed, create new fragments. */ 7256 if (action == PF_PASS && *m0 && (pflags & PFIL_FWD) && 7257 (mtag = m_tag_find(m, PF_REASSEMBLED, NULL)) != NULL) 7258 action = pf_refragment6(ifp, m0, mtag); 7259 7260 SDT_PROBE4(pf, ip, test6, done, action, reason, r, s); 7261 7262 return (action); 7263 } 7264 #endif /* INET6 */ 7265